Attribute sampling calculator: sample size table and audit example

Attribute sampling is the statistical method for tests of controls: it measures how often a control fails, not how much money is wrong. At 95% confidence, a 5% tolerable deviation rate and a 1% expected rate, the sample is 93 items with 1 deviation allowed.

Updated 10 November 2026

Attribute sample size calculator

Test of controls: sample size

C-2.1
Deviations allowed1
DistributionBinomial
Sample size93 items

If the true rate were 5%, the chance of finding 1 deviation or fewer in 93 items would be 4.998% (risk of overreliance, at most 5%).

Evaluate the sample

Upper deviation limit at the same confidence level

C-2.2
Sample deviation rate1.08%
Upper deviation limit5.00%

Upper limit 5.00% ≤ 5% tolerable: the sample supports reliance on the control.

Attribute sampling for tests of controls

In attribute sampling each item in the sample gets one of two answers: the control worked, or it did not. You are counting an attribute, not measuring an amount. From that count you estimate the deviation rate in the population and, more usefully, an upper limit: the rate the population does not exceed, at the confidence level you chose.

It is the method for tests of controls, where the auditor wants to rely on a control (purchase order approval, the monthly bank reconciliation, matching invoices to goods received notes) to cut back substantive testing. If the aim is to quantify misstatement in currency, use monetary unit sampling or stratified variables sampling instead. The full comparison is in attribute vs variable sampling.

ISA 530 (AU-C 530 in the US) prescribes no table and no formula. It asks for a sample size that reduces sampling risk to an acceptably low level. The tables most firms use are the ones in the AICPA Audit Sampling Guide, which come from the exact binomial distribution. That is what the calculator on this page computes.

Defining a deviation

Write down what counts as a deviation before you calculate anything, because everything else depends on it. A deviation is an item where the prescribed control was not applied as designed. For a purchase order approval control it might be:

  • No evidence of approval. No signature, electronic or on paper.
  • Approval above the approver's limit. Signed by someone without authority for that amount.
  • Late approval. Approved after the goods were received.
  • Missing second signature where the policy requires one.

Two rules worth recording in the working paper. If you cannot apply the procedure to a selected item (the purchase order cannot be found, say), that item is a deviation: ISA 530 paragraph 11 requires it. And an error unrelated to the control under test, such as a properly approved order with a mistyped price, is not a deviation from that control, although it still needs following up.

How the attribute sample size is calculated

You need three inputs, all matters of judgement:

  • Confidence level. 95% is usual when the control is an important source of evidence, 90% when you rely on it less. The complement (5% or 10%) is the risk of overreliance: concluding the control works when it does not.
  • Tolerable deviation rate (TDR). The highest rate at which you would still rely on the control. For substantial reliance it is commonly between 3% and 7%.
  • Expected deviation rate (EDR). What you expect to find, from last year or a pilot sample. It has to be clearly below the tolerable rate.

The sample size is the smallest n such that, if the true rate were exactly the tolerable rate, the chance of finding k or fewer deviations is no more than the risk you accept. k, the deviations allowed, is the expected rate applied to the sample and rounded up:

P(X ≤ k | n, TDR) ≤ 1 − confidence, where k = ⌈EDR × n⌉

At 95%, TDR 5% and EDR 1%: with 93 items k = ⌈0.93⌉ = 1, and the chance of seeing 1 deviation or fewer if the true rate were 5% is 4.998%, just under 5%. With 92 items it is not. The shortcut n = factor ÷ (TDR − EDR) found in many textbooks gives different answers, so do not use it if you want to match the tables.

Attribute sample size table at 95% confidence (AICPA Table A-1)

Sample size with the deviations allowed in brackets. Columns: tolerable deviation rate. Rows: expected deviation rate. Large population. Computed with the AuditSampler engine; it matches Table A-1 of the AICPA Audit Sampling Guide.

Expected rate3%4%5%6%7%8%10%
0.00%99 (0)74 (0)59 (0)49 (0)42 (0)36 (0)29 (0)
0.25%157 (1)117 (1)93 (1)78 (1)66 (1)58 (1)46 (1)
0.50%157 (1)117 (1)93 (1)78 (1)66 (1)58 (1)46 (1)
0.75%208 (2)117 (1)93 (1)78 (1)66 (1)58 (1)46 (1)
1.00%257 (3)156 (2)93 (1)78 (1)66 (1)58 (1)46 (1)
1.25%303 (4)156 (2)124 (2)78 (1)66 (1)58 (1)46 (1)
1.50%392 (6)192 (3)124 (2)103 (2)66 (1)58 (1)46 (1)
1.75%562 (10)227 (4)153 (3)103 (2)88 (2)77 (2)46 (1)
2.00%846 (17)294 (6)181 (4)127 (3)88 (2)77 (2)46 (1)
2.50%3,240 (81)513 (13)234 (6)150 (4)109 (3)77 (2)61 (2)
3.00%—1,098 (33)361 (11)195 (6)129 (4)95 (3)61 (2)
4.00%——1,348 (54)421 (17)221 (9)146 (6)89 (4)
5.00%———1,580 (79)478 (24)240 (12)116 (6)

A sample of more than a few hundred items means that, at that expected rate, the test of controls does not pay: it is usually better not to rely on the control and to extend substantive testing.

Attribute sample size table at 90% confidence (AICPA Table A-2)

Same layout with a 10% risk of overreliance (AICPA Table A-2).

Expected rate3%4%5%6%7%8%10%
0.00%76 (0)57 (0)45 (0)38 (0)32 (0)28 (0)22 (0)
0.25%129 (1)96 (1)77 (1)64 (1)55 (1)48 (1)38 (1)
0.50%129 (1)96 (1)77 (1)64 (1)55 (1)48 (1)38 (1)
0.75%129 (1)96 (1)77 (1)64 (1)55 (1)48 (1)38 (1)
1.00%176 (2)96 (1)77 (1)64 (1)55 (1)48 (1)38 (1)
1.25%221 (3)132 (2)77 (1)64 (1)55 (1)48 (1)38 (1)
1.50%265 (4)132 (2)105 (2)64 (1)55 (1)48 (1)38 (1)
1.75%390 (7)166 (3)105 (2)88 (2)55 (1)48 (1)38 (1)
2.00%590 (12)198 (4)132 (3)88 (2)75 (2)48 (1)38 (1)
2.50%2,079 (52)353 (9)158 (4)110 (3)75 (2)65 (2)38 (1)
3.00%—730 (22)258 (8)132 (4)94 (3)65 (2)52 (2)
4.00%——873 (35)274 (11)149 (6)98 (4)65 (3)
5.00%———1,019 (51)318 (16)160 (8)78 (4)

Small populations

The tables assume a large population, and above a few thousand items population size has almost no effect (ISA 530 Appendix 2 says as much). When the sample would be more than a tenth of the population, the calculator switches to the hypergeometric distribution for that exact population, which gives a slightly smaller sample. At 95%, 5% and 1%:

Population (items)SampleDeviations allowedDistribution
2,500931Binomial
1,000931Binomial
500871Hypergeometric
300831Hypergeometric
200781Hypergeometric
100651Hypergeometric
50431Hypergeometric

For very small populations, such as a monthly control (12 performances a year) or a quarterly one, statistical sampling makes little sense: test them all or apply the fixed number in your firm's methodology.

Attribute sampling audit example

Example population: 2,500 purchase orders raised during the year. Control: every order is approved before it is issued by someone with authority for its amount, with a second signature above 50,000. The auditor wants to rely on the control to reduce tests of details on purchases.

  1. Parameters: 95% confidence, 5% tolerable rate, 1% expected rate (last year's sample had one deviation).
  2. Size: 93 orders, 1 deviation allowed. With 2,500 items the population adjustment changes nothing.
  3. Simple random selection of 93 order numbers out of 2,500, keeping the seed so the reviewer can repeat it. The options are covered in audit sample selection methods.
  4. Testing: for each order check the signature, the approver's limit, the approval date against the receipt date and the second signature.
  5. Result: 2 deviations, one order approved twelve days after the goods arrived and one approved above the approver's limit.

Evaluation

The sample deviation rate is 2 ÷ 93 = 2.15%, but that is not the figure that decides. What you compare with the tolerable rate is the 95% upper deviation limit, the exact Clopper-Pearson bound that the AICPA evaluation Tables A-3 and A-4 are built on. For 93 items it moves like this:

Deviations foundSample rateUpper limit (95%)Conclusion at 5% tolerable
00.00%3.17%Reliance supported
11.08%5.00%Reliance supported
22.15%6.62%Reliance not supported
33.23%8.13%Reliance not supported

With 2 deviations the upper limit is 6.62%, above the 5% tolerable rate: the sample does not support the planned reliance on the control. With 1 deviation the limit is 4.9994%, just inside (rounded up to 5.00% in the table), which is why the table allows exactly 1. No projection step is needed for a test of controls: the sample rate already is the projected rate (ISA 530, A20).

When the upper limit exceeds the tolerable rate (ISA 530, A21 y A23): investigate the nature and cause of the deviations, raise the assessed risk of material misstatement and extend substantive procedures, test a compensating control, or extend the sample. If you extend, the whole sample has to pass: with 2 deviations you would need 124 items with no further deviations (limit 4.990%), so 31 more orders.

Counting is not enough. Two deviations by the same person or in the same month point to a specific problem the upper limit cannot see. ISA 530 paragraph 12 requires you to investigate the nature and cause of every deviation, and an intentional one is a fraud indicator, not just another count.

Common mistakes

  • Comparing the sample rate with the tolerable rate. 2.15% is below 5%, but the upper limit is 6.62%. The conclusion rests on the limit.
  • Setting the expected rate too close to the tolerable rate. At TDR 5% and EDR 3% the sample jumps to 361 items; at EDR 4%, to 1,348. If you expect that many deviations, the control will not reduce your testing.
  • Using the shortcut formula instead of the exact binomial and not matching the AICPA tables or what the reviewer expects.
  • Defining a deviation after seeing the results. Write it down before testing.
  • Dropping items you cannot test. A missing document counts as a deviation.
  • Sampling an automated control. An application control that behaves the same for every transaction is tested differently: IT general controls and a test of one.
  • Sampling only part of the period when you intend to rely on the control for the whole year.

Attribute sampling in AuditSampler

AuditSampler Pro sizes the sample with the same binomial rule (hypergeometric for small populations), selects a random sample with a reproducible seed from the population file, records each item as pass or deviation and evaluates the upper limit. The working paper comes out as a PDF.

Frequently asked questions

What is attribute sampling in audit?

The statistical sampling method used for tests of controls. Each item is marked as correct or as a deviation, and the number of deviations gives an upper limit on the population deviation rate, which is compared with the tolerable rate.

How do you calculate an attribute sample size?

It is the smallest number of items for which, if the true rate were the tolerable rate, the chance of finding the allowed number of deviations or fewer is no more than the accepted risk. At 95% confidence, a 5% tolerable rate and a 1% expected rate, that is 93 items with 1 deviation allowed.

What is the tolerable deviation rate?

The highest rate of control failures at which the auditor would still rely on the control. ISA 530 defines it and the auditor sets it according to how much substantive testing the control is meant to replace; for substantial reliance it is commonly 3% to 7%.

How many items do I test if I expect no deviations?

At 95% confidence, 59 items for a 5% tolerable rate, 29 for 10% and 99 for 3%. At 90% confidence, 45, 22 and 76.

Does population size affect attribute sample size?

Hardly, for large populations. Only when the sample is more than a tenth of the population is the hypergeometric calculation worth using: at 95%, 5% and 1%, a population of 500 gives a sample of 87 and a population of 200 gives 78.

What if I find more deviations than expected?

If the upper limit exceeds the tolerable rate, the sample does not support reliance on the control. Investigate the cause, raise the assessed risk and extend substantive testing, test a compensating control, or extend the sample and recalculate with all the deviations found.

  • AICPA, Audit Guide: Audit Sampling (Tables A-1 to A-4)
  • IAASB, ISA 530 Audit Sampling (paragraphs 5, 11, 12 and A20 to A23; Appendix 2)
  • AICPA, AU-C Section 530 Audit Sampling
  • Clopper, C. J. and Pearson, E. S. (1934), The use of confidence or fiducial limits illustrated in the case of the binomial, Biometrika 26

From the table to the working paper

AuditSampler Pro sizes the attribute sample, selects it from the population file with a reproducible seed, records each deviation and evaluates the upper limit in a PDF working paper. Client data stays on your computer.