How do I set up a stability study — how many batches, which conditions, how many time points?

Before you copy a protocol, find out whether it was written for you. If the protocol in front of you carries three batches, 25 °C and 40 °C, it came from ICH Q1A(R2), and that guideline says in its own first paragraph what it is for: it defines the stability data package for a new drug substance or drug product that is sufficient for a registration application within the three regions of the EC, Japan and the United States (§1.1). It addresses the information to be submitted in registration applications for new molecular entities and associated drug products, and it does not currently seek to cover the information to be submitted for abbreviated or abridged applications, variations or clinical trial applications (§1.2). If you are not filing one of those, its numbers are a benchmark you may choose to meet. They are not a rule you are under.

What reaches you, in one paragraph. Part 211 carries the federal stability requirement, and it states its own scope as the minimum current good manufacturing practice for preparation of drug products, excluding positron emission tomography drugs and medical gases, for administration to humans or animals (21 CFR 211.1(a)). The same section then lifts the part off one class: pending consideration of a proposed exemption published in the Federal Register of September 29, 1978, the requirements in the part shall not be enforced for over-the-counter drug products if the products and all their ingredients are ordinarily marketed and consumed as human foods, and parts 110 and 117 are applied to them instead (211.1(c)). Part 111 applies, except as its own paragraph (b) provides for holding at a retail establishment, if you manufacture, package, label or hold a dietary supplement (21 CFR 111.1(a)), and it has no stability-testing section and no expiration-dating section; it names shelf life in three record-retention clauses and conditions every one of them on shelf life dating being used at all (111.83(b)(3), 111.465(b), 111.605(a)). We read every section of 21 CFR parts 700 and 701 and of the cosmetic good manufacturing practice standard ISO 22716 against the words stability, shelf life and expiration, and found no such requirement; the same reading of 21 CFR part 117 returns none for conventional food. Which rule reaches which product, exemption by exemption, is answered in a companion piece rather than repeated here: do I need a real-time stability study, or is accelerated data enough?

What this page does not reach. Medical devices, biologics, animal food and tobacco each run under their own parts and none of them is covered here. Neither is state law, nor the law of any market outside the United States, nor the terms of your own customer agreements, which are free to ask for more than any rule does. And this page does not tell you what your shelf life should be. It tells you what a study capable of supporting one is made of, and where each number in it comes from.

Four terms do the work below, and ICH Q1A(R2) defines each of them in its own glossary rather than leaving them to the reader.

• A primary batch is a batch of a drug substance or drug product used in a formal stability study, from which stability data are submitted in a registration application for the purpose of establishing a re-test period or shelf life. For a drug product, two of the three batches should be at least pilot scale, and the third can be smaller if it is representative with regard to the critical manufacturing steps; a primary batch may also be a production batch (glossary, §3).
• A pilot scale batch is one manufactured by a procedure fully representative of and simulating that to be applied to a full production scale batch. For solid oral dosage forms, a pilot scale is generally at a minimum one-tenth that of a full production scale, or 100,000 tablets or capsules, whichever is the larger (glossary, §3).
• Formal stability studies are the long term and accelerated, and intermediate, studies undertaken on primary or commitment batches according to a prescribed stability protocol, to establish or confirm the re-test period of a drug substance or the shelf life of a drug product (glossary, §3).
• Accelerated testing means studies designed to increase the rate of chemical degradation or physical change by using exaggerated storage conditions as part of the formal stability studies. The guideline attaches a warning to its own definition: results from accelerated testing studies are not always predictive of physical changes (glossary, §3).

On this page: Where your numbers come from · The federal drug rule names no number · How many batches, and why three · Which conditions · How many time points · The decisions that come before the numbers · More than one strength or pack size · What actually gets written up · Five things to settle first

Where your numbers come from

A stability protocol has three numbers a reader will always ask about: how many batches went on test, at what conditions, and at which pull points. Every one of them is set by a different document depending on what your product is, and for several classes of product no document sets them at all.

Where a stability study’s design numbers come from, by what the product is under United States federal law. Provisions read from our own copies on August 18, 2026.

What you make	What sets the design	What it actually specifies
A new drug substance or product going into a registration application	ICH Q1A(R2)	At least three primary batches; long term at 25 °C / 60 % RH or 30 °C / 65 % RH and accelerated at 40 °C / 75 % RH for the general case; for a proposed shelf life of at least 12 months, long term testing normally every 3 months in year one, every 6 months in year two, annually after that; a minimum of three accelerated time points
Any other drug product, including an over-the-counter monograph drug	21 CFR 211.166, subject to the exemptions in 211.137 and in 211.166(c) and (d)	No batch number, no temperature and no pull point. It asks for a written program including sample size and test intervals based on statistical criteria, and for an adequate number of batches. Homeopathic products run under a different written assessment; allergenic extracts labeled “No U.S. Standard of Potency” sit outside the section entirely
A dietary supplement	No federal rule. NSF/ANSI 455-2 where you certify to it	That the shelf life behind any printed date be supported by data. The design itself is written as a recommendation, with ICH Q1 offered as an example
A dietary supplement in the NSF dietary supplements GMP registration program	That program’s own requirements	Where the regulations of the country of manufacture and distribution require a stability program: three batches of the same formulation, real-time pull points defined through the proposed shelf life, and a minimum of three accelerated points including initial and final, unless alternative methods are justified
An over-the-counter drug under the NSF OTC drug GMP standard	NSF/ANSI 455-4	At least three batches in the commercial container closure system; the same 3 / 6 / annual long term schedule; a minimum of three accelerated points
A cosmetic, or a conventional food that is not infant formula	Nothing federal that we found	Nothing. The design is set by your own risk judgment, your certification scheme if you hold one, and your customer

What this table does not tell you. It is a map of which document speaks to which product, not a determination of what your product is. A product can sit in two rows at once: a sunscreen is a cosmetic and a drug, and it is governed as both. It also says nothing about whether your date is right. A design copied correctly out of the right row still produces a wrong shelf life if it never tests the attribute your product actually fails on.

The federal drug rule names no number

This surprises people who arrive expecting the regulation to be the strict version and the guideline to be the soft one. It is the other way around. Section 211.166 requires a written testing program designed to assess the stability characteristics of drug products, requires that the results be used in determining appropriate storage conditions and expiration dates, and requires that the written program be followed (21 CFR 211.166(a)). Then it lists what the program shall include, and the list is part of the requirement:

• sample size and test intervals based on statistical criteria for each attribute examined, to assure valid estimates of stability (211.166(a)(1));
• storage conditions for samples retained for testing (211.166(a)(2));
• reliable, meaningful and specific test methods (211.166(a)(3));
• testing of the drug product in the same container-closure system as that in which it is marketed (211.166(a)(4));
• testing of drug products for reconstitution at the time of dispensing, as directed in the labeling, as well as after they are reconstituted (211.166(a)(5)).

Read (211.166(a)(1)) again. It is the clause that reaches your pull points, and it does not give you any: it asks for a statistical basis for the sample size and the intervals. A schedule copied from a guideline is not a statistical basis, and neither is a schedule copied from the last product you dated. On batches the rule is equally open: an adequate number of batches of each drug product shall be tested to determine an appropriate expiration date, and a record of such data shall be maintained (211.166(b)). Adequate to what, the clause leaves to you and to whoever reads your file. The expiration date itself is then tied back to that testing: to assure that a drug product meets applicable standards of identity, strength, quality and purity at the time of use, it shall bear an expiration date determined by appropriate stability testing described in 211.166 (211.137(a)).

Four classes are lifted out of this, and one of them is large. Homeopathic drug products are exempt from the expiration-dating section (211.137(e)) and are held instead to a written assessment of stability based at least on testing or examination of the product for compatibility of the ingredients, and on marketing experience indicating no degradation over the normal or expected period of use, evaluated in the same container-closure system in which the product is marketed (211.166(c), 211.166(c)(1), 211.166(c)(2)). Allergenic extracts labeled “No U.S. Standard of Potency” are exempt from both sections (211.137(f), 211.166(d)). New drug products for investigational use are exempt from the expiration-dating requirements, provided they meet appropriate standards or specifications as demonstrated by stability studies during their use in clinical investigations (211.137(g)). And pending consideration of that same 1978 proposed exemption, those requirements shall not be enforced for human over-the-counter drug products if their labeling does not bear dosage limitations and they are stable for at least three years as supported by appropriate stability data (211.137(h)). Read that last one to its end. The condition is holding stability data showing three years. It relieves the printed date, not the data behind it, and it says nothing about 211.166.

The section names test intervals once more, and no more helpfully than the first time: where accelerated data has projected a tentative expiration date beyond a date supported by actual shelf life studies, there must be stability studies conducted, including drug product testing at appropriate intervals, until the tentative date is verified or the appropriate date determined (211.166(b)). Appropriate to what, again, is yours to justify.

So the regulation asks you to justify a design. The guideline offers one that has already been justified, for a purpose that may not be yours. That is the whole shape of the problem, and the rest of this page is what is inside the guideline’s answer so you can tell which parts of it your product needs.

How many batches, and why three

ICH Q1A(R2) says data should be provided on at least three primary batches of the drug product, and it puts five conditions on those batches in the same paragraph (§2.2.3). They should be of the same formulation and packaged in the same container closure system as proposed for marketing. The manufacturing process used should simulate that to be applied to production batches, and should provide product of the same quality and meeting the same specification as that intended for marketing. Two of the three should be at least pilot scale batches, and the third can be smaller if justified. Where possible, the batches should be made using different batches of the drug substance. And studies should be performed on each individual strength and container size unless bracketing or matrixing is applied.

The last two are the ones that get dropped in the retelling. Using different batches of active is what turns three batches into evidence about a supply chain rather than three copies of one good day. And “each individual strength and container size” is why a firm with nine SKUs discovers that three batches was never the number.

The reason for three is stated plainly, and it is worth having in your own words because you will be asked for it. The purpose of the study is to establish, based on testing a minimum of three batches, a shelf life and label storage instructions applicable to all future batches manufactured and packaged under similar circumstances; the degree of variability of individual batches affects the confidence that a future production batch will remain within specification throughout its shelf life (§2.2.9). One batch tells you what that batch did. Three tell you something, weakly, about the spread. That is the entire argument, and it is an argument about confidence rather than a magic number.

The guideline also anticipates the case where you do not have three production batches yet. Where the available long term data do not cover the proposed shelf life at approval, a commitment is made to continue the studies afterwards, and the guideline sets out three versions of that commitment depending on how many production batches the submission contained; where the submission already includes long term data from three production batches covering the proposed shelf life, it considers a post-approval commitment unnecessary (§2.2.8).

Two voluntary schemes pick the number up, one flatly and one conditionally. Under the over-the-counter drug good manufacturing practice standard NSF/ANSI 455-4, at least three batches of a product using the same formulation and commercial container closure system shall be selected for stability testing (NSF/ANSI 455-4 §4.6.18.1), and all products bear an expiration date supported by stability data generated under GMP or ICH requirements, except as allowed by the regulations of the country of sale (NSF/ANSI 455-4 §4.6.17). The NSF dietary supplements GMP registration program carries the same three-batch figure in its laboratory controls requirements, and the conditions above it decide whether the figure reaches you at all. That program requires a stability testing program to support product shelf life where required by the regulations of the country of manufacture and distribution; it requires procedures for that program if applicable; and only then does it require the stability testing procedures to include, unless alternative methods are justified, three batches of a product using the same formulation, real-time pull points defined through the proposed shelf life, and a minimum of three accelerated points including initial and final (NSF Dietary Supplements Program GMP requirements, laboratory controls, stability testing procedures). So the three is conditional before it is conditional again, and the first condition is the one this page has already answered: no United States federal regulation requires a stability program for a dietary supplement. Where the number binds flatly is the over-the-counter drug standard, which states that stability data is required for all drug products even where the product is exempt from printing a specific expiration date (NSF/ANSI 455-4 §4.6.17.1). Both are scheme requirements rather than regulations, and each binds you because you signed up to it.

Which conditions

The principle comes before the table. A drug product should be evaluated under storage conditions that test its thermal stability and, if applicable, its sensitivity to moisture or potential for solvent loss, and the conditions and the lengths of studies chosen should be sufficient to cover storage, shipment and subsequent use (§2.2.7). Long term testing should cover a minimum of 12 months on at least three primary batches at the time of submission, and should continue for a period sufficient to cover the proposed shelf life (§2.2.7). Alternative storage conditions can be used if justified.

For the general case, the guideline sets long term at 25 °C ± 2 °C / 60 % RH ± 5 % RH or at 30 °C ± 2 °C / 65 % RH ± 5 % RH, with 12 months of data at submission; intermediate at 30 °C ± 2 °C / 65 % RH ± 5 % RH for 6 months; and accelerated at 40 °C ± 2 °C / 75 % RH ± 5 % RH for 6 months (§2.2.7.1). Two conditions on that are dropped so often that the three-condition design has become the thing people believe the guideline requires. It is up to the applicant which of the two long term conditions to run. And if 30 °C / 65 % RH is the long term condition, there is no intermediate condition at all. The intermediate arm exists only where long term is run at 25 °C / 60 % RH and significant change occurs at any time during the six months at the accelerated condition; then a minimum of 6 months’ data from a 12-month intermediate study goes into the application (§2.2.7.1).

Which makes “significant change” the trigger for a whole additional arm of the study, so it is worth reading to the end. In general, significant change for a drug product is defined as a 5 % change in assay from its initial value, or failure to meet the acceptance criteria for potency where biological or immunological procedures are used; any degradation product exceeding its acceptance criterion; failure to meet the acceptance criteria for appearance, physical attributes and functionality test, with the note that some changes in physical attributes, such as softening of suppositories or melting of creams, may be expected under accelerated conditions; and then, as appropriate for the dosage form, failure to meet the acceptance criterion for pH, or failure to meet the acceptance criteria for dissolution for 12 dosage units (§2.2.7.1). Five items, not three, and the last two carry a qualifier that decides whether they apply to you at all.

The general case is one of six. Its 40 °C / 75 % RH accelerated condition is available in two of the six: the general case itself, and a product in an impermeable container, which the guideline lets you study under any controlled or ambient humidity. In the other four it is a different condition, or none at all.

• Impermeable containers. Sensitivity to moisture or solvent loss is not a concern for a product in a container that is a permanent barrier to their passage, so studies can be run under any controlled or ambient humidity condition (§2.2.7.2).
• Semi-permeable containers. An aqueous product in a semi-permeable container is evaluated for water loss as well as everything else. Long term runs at 25 °C / 40 % RH or 30 °C / 35 % RH, and accelerated at 40 °C / not more than 25 % RH. A 5 % loss of water from the initial value after an equivalent of three months at that accelerated condition is a significant change, except for small containers of 1 mL or less and for unit-dose products, where a loss of 5 % or more may be appropriate if justified. A water loss significant change on its own does not trigger the intermediate arm. The guideline also lets you run at a higher relative humidity and calculate the loss at the reference humidity, and prints the ratios for doing it: at 40 °C the loss rate at not more than 25 % RH is the rate measured at 75 % RH multiplied by 3.0 (§2.2.7.3).
• Refrigerated products. Long term at 5 °C ± 3 °C for 12 months, accelerated at 25 °C / 60 % RH for 6 months. Where significant change occurs between 3 and 6 months at the accelerated condition, the shelf life is based on the real time data. Where it occurs inside the first 3 months, the guideline says it is unnecessary to keep testing to 6 months (§2.2.7.4).
• Frozen products. Long term at −20 °C ± 5 °C for 12 months, and there is no accelerated condition. The shelf life rests on real time data, and a single batch is put through an elevated temperature for an appropriate period instead, to address short term excursions (§2.2.7.5). Below −20 °C is case by case (§2.2.7.6).

If your product is a refrigerated liquid and your protocol says 40 °C / 75 % RH, the design was copied rather than chosen, and the copy is visible to anyone who reads it. The drug regulation makes a smaller demand in the same territory and it is the one that gets cited: the written program shall describe the storage conditions for samples retained for testing (211.166(a)(2)).

How many time points

The guideline sets the frequency in one paragraph, and every sentence in it carries a condition (§2.2.6).

For long term studies the frequency should be sufficient to establish the stability profile of the product. For products with a proposed shelf life of at least 12 months, testing at the long term condition should normally be every 3 months over the first year, every 6 months over the second year, and annually thereafter through the proposed shelf life. Those intervals produce the 0, 3, 6, 9, 12, 18, 24 schedule, and the paragraph writes it for a product claiming at least a year. At the accelerated condition, a minimum of three time points including the initial and final, for example 0, 3 and 6 months, from a six-month study is recommended. Where development experience gives you reason to expect the accelerated results to approach the significant change criteria, testing should be increased, either by adding samples at the final time point or by including a fourth time point. And where the intermediate condition is called for because significant change occurred at the accelerated condition, a minimum of four time points including the initial and final, for example 0, 6, 9 and 12 months, from a twelve-month study is recommended.

The word doing the work in that paragraph is “normally”. The schedule is a default for a proposed shelf life of at least twelve months, not a fixed grid, and reduced designs where the testing frequency is cut can be applied if justified (§2.2.6). Where the same schedule appears in a certification standard it hardens: for long term studies, initial testing and then testing every 3 months over the first year, every 6 months over the second year, and annually thereafter through the proposed shelf life (NSF/ANSI 455-4 §4.6.18.5), with a minimum of three accelerated points including initial and final for a six-month study (NSF/ANSI 455-4 §4.6.18 sub-clause 6).

One consequence people meet late: a pull point you did not schedule is a pull point you cannot add afterwards. Samples have to be in the chamber from time zero, in the marketed pack, in enough quantity for every test you might later want. The cheapest hour in a stability program is the one spent deciding how much to put on test.

The decisions that come before the numbers

Batches, conditions and pull points are the visible part of a protocol and the least of it. The guideline puts the reasoning first: the design of the formal studies should be based on knowledge of the behavior and properties of the drug substance, on stability studies of that substance, and on experience gained from clinical formulation studies, and the likely changes on storage and the rationale for the selection of attributes to be tested should be stated (§2.2.1). That sentence is the difference between a study and a schedule.

What you test follows from it. Studies should include testing of those attributes susceptible to change during storage and likely to influence quality, safety or efficacy, covering as appropriate the physical, chemical, biological and microbiological attributes, preservative content such as antioxidant and antimicrobial preservative, and functionality tests such as dose delivery per actuation. Analytical procedures should be fully validated and stability indicating. And a single primary batch should be tested for antimicrobial preservative effectiveness, in addition to preservative content, at the proposed shelf life, for verification purposes, regardless of whether release and shelf life acceptance criteria for preservative content differ (§2.2.5). That last one sits inside a paragraph most readers skim as being about specifications, and like everything else in the guideline it is written as a should rather than a must.

Two more sit alongside it. Testing should be conducted on the dosage form packaged in the container closure system proposed for marketing, including as appropriate any secondary packaging and container label (§2.2.4). The drug regulation asks for less, in a single line: testing of the drug product in the same container-closure system as that in which the drug product is marketed (211.166(a)(4)). The secondary packaging and the label are the guideline’s addition, not the rule’s. And photostability testing should be conducted on at least one primary batch if appropriate (§2.2.2), under conditions set out in a separate guideline of the same family.

This is where a stability program is won or lost, and it is not a numbers question. A marine oil that oxidizes, a botanical extract whose marker compound drifts, a probiotic losing viable count, a tablet whose disintegration slows as it hardens: each of those fails a different way, and a design built for one is blind to the others. Three batches at 40 °C tested for the wrong attribute is an expensive record of nothing.

More than one strength or pack size

The default is one study per combination: studies should be performed on each individual strength and container size unless bracketing or matrixing is applied (§2.2.3). Both of those are defined designs with defined conditions, not permission to test less.

Bracketing, as ICH Q1A(R2) defines it, is a schedule in which only samples on the extremes of certain design factors, such as strength or container size, are tested at all time points as in a full design, on the assumption that the stability of intermediate levels is represented by the extremes (glossary, §3). It is not appropriate if it cannot be shown that the strengths or container sizes chosen are indeed the extremes (ICH Q1D §2.3). Where both container size and fill vary, it should not be assumed that the largest and smallest containers represent the extremes of all packaging configurations; the characteristics to compare include, as appropriate, container wall thickness, closure geometry, surface area to volume ratio, headspace to volume ratio, and water vapor or oxygen permeation rate per dosage unit or unit fill volume (ICH Q1D §2.3.1.2). And if the extremes turn out to differ, the intermediates are treated as no more stable than the least stable extreme (ICH Q1D §2.3.2).

Matrixing, in the same glossary, tests a selected subset of the possible samples at each time point, a different subset at the next, on the assumption that each subset represents all of them (glossary, §3). It carries conditions of its own. In a design where the time points are matrixed, all selected factor combinations should be tested at the initial and final time points, and where full long term data for the proposed shelf life will not be available before approval, at 12 months or the last time point before submission as well. Data from at least three time points, including initial, should be available for each combination through the first 12 months. And at an accelerated or intermediate condition, care should be taken that testing occurs at a minimum of three time points, including initial and final, for each combination (ICH Q1D §2.4.2). Whether it may be used at all turns on variability: with small variability in the supporting data a matrixing design is applicable; with moderate variability it should be statistically justified; with large variability it should not be applied (ICH Q1D §2.4.4).

Both are reduced designs, and the guideline states the price before the benefit. A reduced design should be able to adequately predict the shelf life, and the risk of establishing a shorter shelf life than a full design would have supported should be considered before it is chosen (ICH Q1D §2.1). A matrixing design on factors other than time points generally has less precision and yields a shorter shelf life than the corresponding full design, though one that matrixes on time points only would often do as well as a full design at detecting differences in rates of change and establishing a reliable shelf life (ICH Q1D §2.4.5). So the honest framing is not that a reduced design saves money for free. It trades data for money, and the guideline asks you to weigh the shorter date before you choose it.

What actually gets written up

The design clauses of the drug rule are cited on inspection, and the wording of those citations is the most useful thing about them, because it is the sentence a reader would write about your protocol. The design clause carries 42 citations of (211.166(a)(1)) across 39 firms, fiscal years 2009 through 2026, and they are written three ways. Twenty-two say the written stability program for drug products does not include test intervals based on statistical criteria for each attribute examined to assure valid estimates of stability. Fifteen say it does not include sample size. Five say it does not include sample size and test intervals. The written stability program does not assure testing of the drug product in the same container-closure system as that in which the drug product is marketed: 40 citations of (211.166(a)(4)) across 37 firms. The written stability program for drug products does not describe the storage conditions for samples retained for testing: 41 citations of (211.166(a)(2)) across 40 firms.

What those figures do not evidence. They are counts of clause citations in published inspection observations, one row per citation, not counts of firms, inspections, warning letters or enforcement actions, and one inspection can contribute several rows. Every one of them sits in the Drugs program area, because the clause they were written under reaches drug products and nothing else; none of it says anything about a supplement, cosmetic or food maker. They are also small numbers next to the rest of the section, and the reason is not that designs are usually sound. The full breakdown across 211.166 and 211.137, including the far larger count for programs that were absent or not followed, is set out in the companion piece on accelerated and real-time data and is not repeated here. Fiscal year 2026 was still open at the read date, so its share is partial.

Five things to settle before you write the protocol

Take dated copies of anything you already hold before you revise it. A stability protocol and an expiration-dating justification are controlled records, and changing one is your own quality unit’s decision rather than a tidy-up.

1. Settle your product class, then find your row. The answer for a drug product and the answer for a dietary supplement are not the same answer, and a product that is both is governed as both. Everything below depends on this and nothing above it does.
2. Name the failure mode before the conditions. Write down how you expect this product to go off, and which measured attribute would show it. If that sentence is hard to write, the protocol built without it will be harder to defend.
3. Check the package in the protocol against the package on the truck. Same container, same closure, same fill, same secondary packaging where it matters. This is the design error that survives every review until someone compares two documents.
4. Count your combinations, not your batches. Strengths multiplied by container sizes multiplied by fills. If the number is uncomfortable, bracketing and matrixing are the named ways to reduce it, each with its own conditions and each costing you shelf life.
5. Decide the sample quantity for tests you have not scheduled yet. You cannot add a pull point or an assay retrospectively to material that was never put in the chamber. Overfill the study now or accept the design you have.