Rule 81 of Income explained: this guide covers what it means, who it applies to, the step-by-step process, documents required, fees, due dates and penalties in India — so you can stay compliant with confidence and avoid costly mistakes.
Rule 81 of the Income-tax Rules, 2026 applies where the most appropriate method produces more than one price. A dataset is built in ascending order; where it holds six or more entries and the method is CUP, resale price, cost plus or transactional net margin, the arm's length range runs from the 35th to the 65th percentile, and a price outside it is replaced by the median. Otherwise the arithmetical mean applies, with a tolerance not exceeding 3%.
When rule 81 is triggered — sub-rules (1) and (2)
Rule 81 applies where application of the most appropriate method referred to in section 165(3)(b) results in determination of more than one price. A dataset is then constructed by placing those prices in ascending order, and the arm's length price is determined on that dataset.
Multi-year data — sub-rules (3), (4) and (5)
Where the most appropriate method is the resale price method, cost plus method or transactional net margin method, and the comparable uncontrolled transaction of a third-party enterprise has been identified using data relating to:
| Clause | Data used | Consequence |
|---|---|---|
| (a) | The current year, and that enterprise undertook the same or similar comparable uncontrolled transaction in either or both of the two financial years immediately preceding | Determine the price for those years by applying the most appropriate method in the same manner as in the current year, and include the weighted average of those prices in the dataset instead of the single price |
| (b) | The financial year immediately preceding the current year, where current-year data is not available at the time of furnishing the return, and that enterprise undertook the same or similar transaction in the financial year immediately preceding those two financial years | Determine the price on the same basis as applied for the immediately preceding year, and include the weighted average in the dataset instead of the single price |
The exclusion in sub-rule (4)
Where the use of current-year data in terms of rule 79(5) establishes that:
- (a) the enterprise has not undertaken the same or similar uncontrolled transaction during the current year; or
- (b) the uncontrolled transaction undertaken during the current year is not a comparable uncontrolled transaction,
then, irrespective of anything else, neither the price nor the weighted average of prices of the comparable uncontrolled transactions shall be included in the dataset.
This is not a downweighting. If current-year testing shows the comparable did not transact this year, or that what it did is not comparable, its earlier years' data cannot be used either — the whole entry leaves the dataset. A study that keeps a comparable alive on two-year-old data after the current year fails it has built a dataset the rule does not permit.
How the weighted average is computed — sub-rule (5)
| Sl. | Method | Factor to which weight is assigned |
|---|---|---|
| 1 | Resale price method | Quantum of sales considered for arriving at the respective prices |
| 2 | Cost plus method | Quantum of costs considered for arriving at the respective prices |
| 3 | Transactional net margin method | Quantum of costs incurred, sales effected, assets employed or to be employed, or any other base considered for arriving at the respective prices |
The weight always tracks the denominator of the profit level indicator. A margin on operating expense is weighted by operating expense; a margin on sales by sales. A simple arithmetic average across years is not a weighted average and does not satisfy sub-rule (5).
The arm's length range — sub-rule (6)
Where the most appropriate method is the comparable uncontrolled price method, resale price method, cost plus method or transactional net margin method, and the dataset consists of six or more entries:
- an arm's length range is constructed beginning from the 35th percentile of the dataset and ending on the 65th percentile;
- (a) if the actual transaction price is within the range, that price is deemed to be the arm's length price;
- (b) if it is outside the range, the median of the dataset is used to compute the arm's length price.
The range applies only where both conditions hold — one of the four named methods and six or more dataset entries. The profit split method and the rule 78 other method are outside sub-rule (6) entirely, however large the dataset. And a dataset of five entries falls to sub-rule (7) even where the method qualifies. Counting the entries is therefore a substantive step, not housekeeping.
Note the asymmetry in clause (b): being outside the range does not move the price to the nearest boundary of the range. It moves to the median — the centre. An adjustment for a price just below the 35th percentile therefore lands well inside the range, not at its edge.
Where the range does not apply — sub-rule (7)
The arm's length price is then:
- (a) the arithmetical mean of all the values included in the dataset; or
- (b) the price at which the transaction has actually been undertaken, if the variation between the mean-based arm's length price and the actual price does not exceed such percentage, not exceeding 3% of the actual price, as may be notified by the Central Government.
The tolerance is a ceiling, not a rate: the rule caps the notifiable percentage at 3%, and the operative figure is whatever the Central Government notifies within that cap. Check the notification for the relevant tax year rather than assuming 3%.
The three definitions — sub-rule (8)
All three work on the dataset arranged in ascending order, and all three share the same two-limb structure.
| Term | Definition |
|---|---|
| Median | (i) the lowest value in the dataset where at least 50% of the values are less than or equal to it; or (ii) the arithmetic mean of that lowest value and the value immediately succeeding it, if the number of all values equal to or less than that value is a whole number |
| 35th percentile | (A) the lowest value such that at least 35% of the values are equal to or less than it; or (B) the arithmetic mean of that value and the value immediately succeeding it, if the number of all values equal to or less than that value is a whole number |
| 65th percentile | (A) the lowest value such that at least 65% of the values are equal to or less than it; or (B) the arithmetic mean of that value and the value immediately succeeding it, if the number of all values equal to or less than that value is a whole number |
These are not the percentile functions built into spreadsheet software, which interpolate. The rule defines a specific selection procedure, and a study that reports a spreadsheet percentile will differ from the rule's answer on many datasets. Compute them as the rule states.
The worked illustration in the rule
Rule 81 carries its own illustration. In it, Enterprise X has undertaken a controlled transaction with its associated enterprises during the current year. The transactional net margin method is the most appropriate method, Enterprise X is the tested party, and the ratio of operating profit to operating expense is the profit level indicator. Current-year data for the comparables is available at the time of furnishing the return, and seven non-associated enterprises have been identified as having undertaken comparable uncontrolled transactions in the current year, all of which also transacted comparably in the two preceding years.
That fact pattern is the standard case for the whole rule working end to end: current-year data available, so sub-rule (3)(a) applies; multi-year comparables, so weighted averages under sub-rule (5) enter the dataset; and seven entries, so the sub-rule (6) range applies. Read the illustration in the book for the arithmetic.
Rule 10CA and rule 81 compared
| Point | Rule 10CA (1962) | Rule 81 (2026) |
|---|---|---|
| Parent provision | Section 92C(2) | Section 165(3)(b) |
| Range | 35th to 65th percentile | 35th to 65th percentile |
| Minimum dataset | Six entries | Six or more entries |
| Eligible methods for the range | CUP, RPM, CPM, TNMM | CUP, RPM, CPM, TNMM |
| Outside the range | Median | Median |
| Fallback | Arithmetical mean, tolerance band | Arithmetical mean, notified percentage not exceeding 3% |
| Multi-year weighting | Sales / costs / relevant base | Sales / costs / relevant base |
Compliance checklist
- Arrange the dataset in ascending order before anything else.
- Count the entries. Six or more opens the range; five closes it.
- Confirm the method is one of the four named — profit split and the rule 78 other method never get the range.
- Apply sub-rule (4) first: drop any comparable that fails the current-year test completely.
- Weight multi-year prices by the denominator of the profit level indicator.
- Compute the median and percentiles as rule 81(8) defines them, not with a spreadsheet percentile function.
- Where the price is outside the range, adjust to the median, not to the nearest boundary.
- Check the notified tolerance percentage for the year; 3% is the statutory ceiling.
Common mistakes
- Using a spreadsheet PERCENTILE function. It interpolates; the rule selects.
- Building the range on five comparables.
- Applying the range to a profit split analysis.
- Adjusting to the 35th percentile rather than the median.
- Taking a plain average across years instead of a weighted average.
- Keeping a comparable on prior-year data after it fails the current-year test.
- Assuming the tolerance is 3% without checking the notification.