Tests of controls sample size by control frequency
The sample size for tests of controls depends on how often the control operates during the period. For infrequent controls, the AICPA Audit Sampling guide suggests 2 instances of a quarterly control, 2 to 4 of a monthly control and 5 to 9 of a weekly control. For daily or recurring controls, firms use samples of 25 to 60, which match a statistical attribute sample with 0 or 1 deviation.
Updated 10 November 2026
Sample size table by control frequency
| Frequency | AICPA, Table 3-5 | Range across firms (FRC, 2023) | KPMG, SOX 404 example (2004) |
|---|---|---|---|
| Annual | – | – | 1 |
| Quarterly | 2 | 1 to 3 | 2 to 3 |
| Monthly | 2 to 4 | 2 to 4 | 2 to 4 |
| Weekly | 5 to 9 | 4 to 11 | 5 to 10 |
| Daily | – | – | 15 to 30 |
| Many times a day | – | 10 to 60 | 30 to 60 |
The three columns don't carry the same weight. Table 3-5 of the AICPA Audit Sampling guide (2017 edition, paragraph 3.62) is non-authoritative guidance for infrequent controls and small populations; its figures are quoted here through the FRC report, because the guide is paid. The UK FRC recorded in 2023 the ranges used by the firms it reviewed. The KPMG table is an example for management's own testing under SOX 404, not for the auditor. No standard sets these numbers: ISA 530 requires a size that reduces sampling risk to an acceptably low level.
Where 25, 40 and 60 come from
For controls that operate hundreds of times, the size comes from attribute sampling. With the AuditSampler Pro engine, for a large population:
| Confidence | Tolerable deviation rate | 0 deviations | 1 deviation | 2 deviations |
|---|---|---|---|---|
| 90 % | 10 % | 22 | 38 | 52 |
| 95 % | 10 % | 29 | 46 | 61 |
| 90 % | 5 % | 45 | 77 | 105 |
| 95 % | 5 % | 59 | 93 | 124 |
Firms' round numbers fit this table. A sample of 25 with no deviations leaves an upper deviation limit of 8.8 % at 90 % confidence (11.3 % at 95 %). With 40 items and one deviation, the limit is 9.4 % at 90 %. With 60 and none, it is 4.9 % at 95 %. That is why the FRC saw two firms move from 25 to 40 when they allow for one expected deviation. The reasoning behind each figure is in the attribute sampling guide.
Why infrequent controls aren't sampled the same way
With small populations, the statistical calculation gets close to testing everything. With no deviations, 90 % confidence and a 10 % tolerable rate, a monthly control (12 instances) would need 8, a weekly control (52) would need 16, and a quarterly control all 4. The AICPA table suggests much smaller samples for these cases, and ISA 530 Appendix 2 accepts that, for small populations, sampling may not be as efficient as other procedures. ISA 330 (paragraph A28) lists the frequency of the control among the factors that drive the extent of testing.
If the control operates once a year, there is no sample: you test that instance. ISA 530 only applies when the auditor decides to use audit sampling (paragraph 1), and the PCAOB says so explicitly for many tests of controls (AS 2315.32), such as segregation of duties.
Automated controls
An automated control does the same thing every time as long as the program doesn't change. If IT general controls (access, change management, operations) are effective, the AICPA guide says testing it once or a few times for each type of transaction is usually enough (paragraphs 1.13 and 1.19), and ISA 330 (A29) accepts not extending the test. If IT general controls fail, a single instance no longer supports the automated control.
What to do when a deviation turns up
- Investigate its nature and cause (ISA 530, paragraph 12): a one-off slip, a stand-in who didn't know the control and a design flaw are treated differently.
- Evaluate with the actual sample. With 25 items and one deviation, the upper limit is 14.7 % at 90 %: above a 10 % tolerable rate, so that size no longer supports reliance on the control.
- Extend or change approach. Extending to 40 only works if no more deviations appear; otherwise reduce reliance on the control and increase substantive testing.
- Document the decision in the working paper: size, items tested, deviations and conclusion. The audit sampling documentation guide sets out what should be on file.
Frequently asked questions
How many instances should I test for a monthly control?
The AICPA Audit Sampling guide table suggests 2 to 4, and the FRC found the same range at the firms it reviewed. A statistical calculation at 90 % confidence with no deviations would need 8 of 12, which is why infrequent controls follow the table.
What sample size is used for a daily control?
The AICPA table gives no figure. In practice firms use 25 to 60 items, which correspond to attribute sampling at 90-95 % confidence with a 5-10 % tolerable rate and 0 or 1 expected deviation.
Why 25 items for tests of controls?
Because 25 items with no deviations leave an upper deviation limit of 8.8 % at 90 % confidence, below a 10 % tolerable rate. If one deviation turns up, the limit rises to 14.7 % and the sample is no longer enough.
Is testing an automated control once enough?
Testing once per type of transaction is usually enough if IT general controls are effective and the program hasn’t changed, according to the AICPA guide and ISA 330 (A29).
Does ISA 530 set minimum sample sizes for tests of controls?
No. It requires a size that reduces sampling risk to an acceptably low level (paragraph 7) and lists the factors that drive it in Appendix 2. Frequency tables are AICPA guidance or each firm’s own practice.
- AICPA, Audit Guide: Audit Sampling (2017), Table 3-5 (paragraph 3.62) and paragraphs 1.13 and 1.19
- FRC, Thematic Review: Audit Sampling (2023)
- KPMG, Sarbanes-Oxley Section 404: Management’s Assessment Process (2004)
- IAASB, Handbook 2023-2024: ISA 330, ISA 530
- IAASB, ISA 530 (paragraphs 1, 7 and 12; Appendix 2) and ISA 330 (A28 and A29)
- PCAOB, AS 2315 Audit Sampling
Size and limit, worked out
AuditSampler Pro works out the attribute sample size and the upper deviation limit with the exact binomial, and produces the working paper as a PDF.