Audit sample selection methods
ISA 530 describes five sample selection methods: random, systematic, monetary unit, haphazard and block selection. The first three work for statistical sampling; haphazard selection only for non-statistical sampling, and block selection rarely supports conclusions about the whole population. Testing all items or selecting specific items are also ways of choosing what to test, but they are not sampling.
Updated 10 November 2026
Audit sampling types and selection methods
ISA 530 requires every sampling unit to have a chance of selection and the sample to be representative (paragraph 8). Appendix 4 describes the methods. In summary:
| Method | How it works | Fit for statistical sampling? | When to use it | Main risk |
|---|---|---|---|---|
| Simple random | Random numbers over the list of items | Yes | Attributes, variables, uniform populations | Not keeping the seed or the population used |
| Systematic | One item every interval, from a random start | Yes, with a random start | Long listings with no pattern | The interval matching a pattern |
| Monetary unit (MUS) | Systematic over cumulative amounts | Yes | Tests of details for overstatement | Negative or zero items never selected |
| Haphazard | Auditor picks without a structured technique | No | Non-statistical sampling on paper records | Bias towards convenient items |
| Block | Contiguous items (a month, a series) | Generally no | As a targeted procedure, not a sample | Concluding on the year from one month |
Audit random sampling
You number the items and a random number generator picks which ones go in. Every item has the same probability of selection. It is the standard method for attribute sampling and variables sampling, and within strata it gives stratified random sampling (see stratified sampling).
What matters is that it can be repeated. Keep the generator's seed and the exact population and anyone gets the same items. For example, from 2,400 invoices numbered 1 to 2,400, a random selection of 10 without replacement with seed 2026 gives invoices 64, 192, 429, 852, 876, 889, 1,120, 1,532, 1,544 and 2,037. Same seed and same list, same result every time.
- When: electronic populations without large differences in amount, or within each stratum.
- Common mistakes: "random" picked by hand with no generator; duplicates with no rule for them; selecting from a listing that later changes.
- Documentation: tool, seed, population file with item count and total, list of selected items.
Systematic selection
Divide the number of items by the sample size to get the interval, choose a random starting point within the first interval and take one item every interval. ISA 530 allows the start to be chosen haphazardly but notes the sample is more likely to be truly random if it comes from a random number generator or table.
Worked example
- Population: 2,400 sales invoices numbered 1 to 2,400. Sample size: 60.
- Interval: 2,400 ÷ 60 = 40.
- Random start between 1 and 40: a generator with seed 530 gives 12.
- Items: 12, 52, 92, 132, 172 and so on up to 2,292, 2,332 and 2,372. Exactly 60 items.
- Pattern check: before accepting the sample, make sure the listing has no 40-item cycle (say, 40 shops invoicing in the same order every day). If it does, every item would come from the same shop.
If the number of items is not a multiple of the sample size, either keep the interval with decimals and round each position, or round the interval down and accept a few extra items. Either way, document the rule.
- When: long listings ordered by date or number, especially when items are pulled from paper files.
- Common mistakes: always starting at item 1; not checking for a cyclical order; cutting the sample short at the end of the listing.
- Documentation: interval, random start and how it was obtained, order of the listing.
Monetary unit (PPS) selection
A systematic selection where the sampling unit is each currency unit. You accumulate the amounts, choose a random start within the first monetary interval and select the item that contains each point. The chance of selection is proportional to size (PPS), so the sample concentrates on large items. The formula is in monetary unit sampling.
Example with ten invoices
Interval 10,000, random start 3,517. The selection points are 3,517, 13,517, 23,517 and so on up to 83,517: nine points over 92,000 of positive amounts.
| Invoice | Amount | Cumulative | Points that fall in it | Result |
|---|---|---|---|---|
| F-101 | 4,200 | 0 – 4,200 | 3,517 | Selected |
| F-102 | 18,750 | 4,200 – 22,950 | 13,517 | Selected (above the interval) |
| F-103 | 950 | 22,950 – 23,900 | 23,517 | Selected |
| F-104 | 7,300 | 23,900 – 31,200 | None | Not selected |
| F-105 | −600 | Not accumulated | None | Excluded: tested separately |
| F-106 | 31,400 | 31,200 – 62,600 | 33,517, 43,517, 53,517 | Selected three times, tested once |
| F-107 | 2,650 | 62,600 – 65,250 | 63,517 | Selected |
| F-108 | 12,100 | 65,250 – 77,350 | 73,517 | Selected (above the interval) |
| F-109 | 5,480 | 77,350 – 82,830 | None | Not selected |
| F-110 | 9,170 | 82,830 – 92,000 | 83,517 | Selected |
Nine points, seven invoices. Those of 10,000 or more (F-102, F-106, F-108) are selected whatever the start. The 950 invoice came up by chance, and F-105, being negative, could never come up: negative and zero balances are set aside and tested another way.
- When: tests of details for overstatement (receivables, inventory, sales) with few expected misstatements.
- Common mistakes: forgetting negative items; counting an item hit three times as three items; using MUS to look for understatement or omissions, which cannot be selected.
- Documentation: interval, random start, order of the listing, and excluded items with their total.
Haphazard selection in audit sampling
The auditor picks items without a structured technique while trying to give every item a chance. ISA 530 asks you to avoid conscious bias or predictability, such as skipping hard-to-find documents or always taking the first or last entry on a page, and states that haphazard selection is not appropriate when using statistical sampling.
- When: non-statistical sampling on physical records, where numbering items and generating random numbers is impractical.
- Common mistakes: drifting towards the largest, most recent or most accessible items; calling a haphazard selection "random".
- Documentation: how items were chosen, from which boxes or months, and why the selection covers the whole population.
Block selection: why it is not a sample
Block selection takes one or more runs of contiguous items, such as all March invoices. The trouble is that neighbouring items resemble each other and not the rest: same staff, same processes, same season. ISA 530 therefore says block selection cannot ordinarily be used in audit sampling and would rarely be appropriate when you want to draw conclusions about the whole population.
Examining a block can still be a good procedure in its own right, for example sales in the last five days of the year for a cut-off test. But then it is a selection of specific items and the result is not projected to the year.
All items or specific items: not sampling
ISA 500 lists two other means of selection. Testing 100% fits when the population has few, large items, when there is a significant risk and other means are not enough, or when a system performs the same calculation repeatedly and data analytics can test it all cheaply. Selecting specific items (the largest, unusual ones, related parties) is often efficient but is not sampling: the result cannot be projected and gives no evidence about the rest of the population.
The usual approach combines both: test the key items 100% and sample the remainder, documenting and evaluating each part separately.
What to document about the selection
- Population file: name, extraction date, item count, total and reconciliation to the ledger.
- Items set aside before selection (key, negative, zero) and their total.
- Method and why it was chosen.
- Seed, or interval and random start, and the tool used.
- List of selected items and any replacement with its reason.
If the sample cannot be reproduced from what is in the working paper, a reviewer cannot check that the selection was random.
Frequently asked questions
What are the types of audit sampling?
By approach, statistical and non-statistical. By selection method, ISA 530 describes random, systematic, monetary unit, haphazard and block selection. Testing all items or selecting specific items is not sampling.
What is systematic selection in audit sampling?
Taking one item every fixed interval (number of items ÷ sample size) from a random starting point within the first interval. With 2,400 items and a sample of 60, the interval is 40.
Is haphazard selection acceptable in audit sampling?
Yes, for non-statistical sampling, provided you avoid bias and every item can be selected. ISA 530 says it is not appropriate when using statistical sampling.
Why is block selection not used for audit sampling?
Contiguous items tend to resemble each other and differ from the rest of the population, so a block is rarely representative. It can be a good targeted procedure, such as a cut-off test, but does not support conclusions about the whole year.
Do I need to keep the seed for random selection?
You should. With the seed, the tool and the exact population, anyone can reproduce the selection and see that it was random, which is what a reviewer will want to check.
Is selecting the largest items audit sampling?
No. It is selecting specific items under ISA 500. It is useful for coverage, but the result is not projected to the rest of the population.
- IAASB, ISA 530 Audit Sampling, Appendix 4
- IAASB, ISA 500 Audit Evidence
- AICPA, Audit Guide: Audit Sampling
Reproducible selection, no spreadsheets
AuditSampler Pro runs seeded random selection and systematic monetary unit selection with a random start, sets negative items aside, and records the seed or start in the PDF working paper.