Audit sample size: how to determine it, with formulas and tables
Audit sample size depends on the risk you accept (confidence), tolerable misstatement or deviation rate, the misstatement you expect to find and, for tests of details, the population value relative to tolerable misstatement. It does not depend on a percentage of the population. With a balance of 2,650,000, tolerable misstatement of 80,000, expected misstatement of 10,000 and 95% confidence, monetary unit sampling gives 125 items; a test of controls at 95% with a 5% tolerable and 1% expected deviation rate gives 93.
Updated 10 November 2026
What ISA 530 says about sample size
ISA 530 (AU-C 530 in the US) gives no formula and no minimum. It requires a sample size sufficient to reduce sampling risk to an acceptably low level, and leaves the choice between a statistical and a non-statistical approach to the auditor. Its Appendices 2 and 3 list the factors that influence sample size for tests of controls and for tests of details, and the standard notes that in similar circumstances those factors affect sample size in the same way whichever approach you take.
In practice, the formulas and tables firms use come from the AICPA Audit Sampling Guide or from firm methodologies built on it. That is the audit sample size guidance most auditors work from.
Factors that drive audit sample size
Each factor pushes the sample one way. The last column shows the effect on this guide's monetary unit sampling example (2,650,000 population, tolerable misstatement 80,000, expected misstatement 10,000, 95%), changing one input at a time:
| Factor | When it increases… | MUS example |
|---|---|---|
| Base case | – | 125 |
| Desired confidence (lower risk of incorrect acceptance) | sample size goes up | 90%: 95 · 99%: 201 |
| Assessed risk of material misstatement | goes up (the test must give more assurance) | – |
| Assurance from other procedures or controls on the same assertion | goes down (less assurance needed) | – |
| Tolerable misstatement | goes down | 120,000: 77 |
| Expected misstatement | goes up, very fast near tolerable | 0: 100 · 20,000: 166 |
| Population value (MUS) | goes up in proportion | 5,300,000: 249 |
| Number of items in the population | barely changes, except for small populations | – |
| Variability of amounts (classical variables) | goes up; stratification brings it down | – |
Key or individually significant items you test 100% come out of the population before you calculate, and their value no longer counts towards sample size.
Audit sample size formula for each method
Monetary unit sampling (tests of details)
n = BV × CF ÷ (TM − EM × EF)
BV is the population book value, CF the confidence factor (3.00 at 95%), TM tolerable misstatement, EM expected misstatement and EF the expansion factor (1.6 at 95%). Example with a receivables balance:
- Expanded expected misstatement: 10,000 × 1.6 = 16,000.
- Room left: 80,000 − 16,000 = 64,000.
- Theoretical interval: 64,000 ÷ 3.00 = 21,333.33.
- Sample size: 2,650,000 ÷ 21,333.33 = 124.2, rounded up to 125 items, with a selection interval of 2,650,000 ÷ 125 = 21,200.
At 90% confidence (CF 2.31, EF 1.5) the same population needs 95 items. Try your own figures in the monetary unit sampling calculator, and see how the results are evaluated in evaluating MUS results.
Attribute sampling (tests of controls)
You set confidence, the tolerable deviation rate and the expected rate. Sample size is the smallest n for which, if the true rate were the tolerable one, the chance of finding no more than the expected deviations would be below the accepted risk. It comes from the binomial distribution and matches the AICPA tables:
| Confidence and tolerable rate | 0% expected | 1% expected | 2% expected |
|---|---|---|---|
| 90%, tolerable 5% | 45 | 77 | 132 |
| 90%, tolerable 10% | 22 | 38 | 38 |
| 95%, tolerable 5% | 59 | 93 | 181 |
| 95%, tolerable 10% | 29 | 46 | 46 |
Large population. Source: AICPA, Audit Guide: Audit Sampling, sample size tables for tests of controls; figures recomputed with the exact binomial distribution.
Example: 4,200 purchase invoices tested for prior approval, at 95% confidence, a 5% tolerable and a 1% expected deviation rate. The sample is 93 invoices, planned to allow one deviation. If you find one, the upper deviation limit is just under 5% (4.9994%) and the control is effective; with two it rises to 6.6% and you cannot rely on it as planned. The attribute sampling calculator runs this with your figures.
Classical variables sampling
This estimates the audited value of the population from the sample mean or the sample differences. Size depends mostly on the population standard deviation and the precision you need (tolerable minus expected misstatement). For a single stratum with N items, standard deviation S, precision A and the z factor for the confidence level:
n = (N × S)² ÷ ((A ÷ z)² + N × S²)
With widely spread amounts the sample balloons, which is why it is nearly always stratified: splitting the population into value bands lowers the spread inside each stratum and the total sample. See stratified sampling.
Non-statistical sampling
A non-statistical sample is not automatically smaller. ISA 530 asks you to consider the same factors and notes that, in similar circumstances, they affect sample size in the same way under either approach. With the same inputs, the size should be close to that of a statistical sample. For tests of details the guide starts from the population value (excluding items tested 100%) divided by tolerable misstatement, multiplied by an assurance factor that rises with the assessed risk of material misstatement and falls when other substantive procedures cover the same assertion.
For tests of controls, many firms use internal tables of minimum sizes by control frequency (annual, quarterly, monthly, weekly, daily, many times a day). Those are firm policy, not the standard: if you use one, cite your methodology's table in the working paper. The sample size table by control frequency compares the AICPA table with firms' ranges.
Why an audit sample size percentage does not work
"Test 10% of invoices" has no link to risk. Once a population is large, the required size hardly depends on the number of items. At 95% confidence, 5% tolerable and 1% expected:
| Items in population | Sample size | 10% of population |
|---|---|---|
| 200 | 78 | 20 |
| 500 | 87 | 50 |
| 1,000 | 93 | 100 |
| 5,000 | 93 | 500 |
| 50,000 | 93 | 5,000 |
For small populations the sample drops a little because it is computed with the hypergeometric distribution. A fixed percentage falls short on small populations and over-tests large ones, and says nothing about sampling risk. MUS works the same way: what matters is the population value relative to tolerable misstatement, not the invoice count.
How to select audit sample size, step by step
- Define the objective of the test and the population, and reconcile it to the ledger.
- Choose the method: attribute sampling for controls; MUS or variables for amounts.
- Take out key or individually significant items to test 100%.
- Set tolerable misstatement (or the tolerable deviation rate) from performance materiality.
- Estimate expected misstatement or deviations from prior years and your knowledge of the client.
- Set confidence from the assessed risk and the assurance other procedures provide.
- Calculate the size and document the inputs, the formula or table used and the result.
Frequently asked questions
How do you determine audit sample size?
From confidence (the sampling risk you accept), tolerable misstatement or deviation rate, expected error, assessed risk and the assurance from other procedures. For tests of details also the population value relative to tolerable misstatement and, for variables sampling, its variability.
What is the audit sample size formula?
It depends on the method. For monetary unit sampling, n = BV × CF ÷ (TM − EM × EF): 2,650,000 with 80,000 tolerable, 10,000 expected and 95% confidence gives 125 items. Attribute samples use the binomial distribution or the AICPA tables.
Is there an audit sample size table?
For tests of controls, yes: the AICPA tables give the size by confidence, tolerable and expected deviation rate (95%, 5% and 1% give 93). For tests of details with MUS you use the formula with the confidence and expansion factors.
Can audit sample size be a percentage of the population?
Not as a valid basis. In large populations the size barely changes with the number of items: at 95% with 5% tolerable and 1% expected it is 93 items for both 1,000 and 50,000. A percentage does not measure sampling risk.
Does ISA 530 set a minimum sample size?
No. It requires a size that reduces sampling risk to an acceptably low level and lists the factors that affect it. Minimums found in some methodologies are firm decisions.
Can a non-statistical sample be smaller?
Not just because it is non-statistical. ISA 530 asks you to weigh the same factors and notes that in similar circumstances they affect size the same way, so with the same inputs the size should be similar.
- IAASB, ISA 530 Audit Sampling (Appendices 2 and 3)
- AICPA, AU-C 530 Audit Sampling
- AICPA, Audit Guide: Audit Sampling
- ICAC, NIA-ES 530 Muestreo de auditoría
Size, selection and evaluation on one working paper
AuditSampler Pro works out monetary unit, attribute and stratified sample sizes, selects the items and records the inputs it used.