DIVIDE() or the / operator in DAX: safe division in Power BI measures

Division is the most common calculation in any business model: average revenue per unit, discount rate, year-over-year growth, conversion rate. It is also the most common source of silent wrong numbers in a Power BI model, because two ways of writing it disagree on exactly one case — the day the denominator is zero.

What the two actually do

DAX gives you two ways to divide:

Revenue Per Unit = [Revenue] / [Units]
Revenue Per Unit = DIVIDE([Revenue], [Units], 0)

When [Units] is a normal, nonzero number, both return the same value. The difference is what each returns when the denominator is zero or blank:

Expression	1 / 0	0 / 0
x / y operator	Infinity	NaN
DIVIDE(x, y, 0)	0	0
DIVIDE(x, y)	BLANK	BLANK

The slash operator never throws an error. A measure that computes Infinity or NaN does not crash the report — it renders. A card shows ∞. A table sorts on a meaningless value. A conditional-formatting rule colors cells against an infinite scale. Everything looks finished; some of the numbers are not.

Why this lands in real models

A 0 denominator is rarely exotic. In a sales model, units can be zero for products that had quotes but no orders; a discount rate divides by a list price that is blank for some rows; a growth measure divides by last year's value, which is zero for new markets. Every one of those is correct data and a formula that only fails in the rows where someone looks.

It also lands through editing. When a report becomes a Power BI project and agents or colleagues refactor the model file by file, a cleanup pass can rewrite a wrapped measure back to a bare slash — the surrounding rows still work, so nobody notices. That workflow problem is bigger than this one rule; we covered it in score your Power BI model before AI agents refactor it.

The fix, and the choice inside it

The mechanical fix is what our audit recommends: replace the slash with DIVIDE(numerator, denominator, 0). The third argument is the alternate result returned on division by zero, and choosing it is the part that deserves a thought:

• DIVIDE(n, d, 0) — right for quantities that genuinely are zero: growth, rates, per-unit prices. A card reads 0 and a chart plots a flat line, which is the honest picture.
• DIVIDE(n, d) — returns BLANK on a zero denominator. Right when "no value" is truer than "zero": a customer with no purchases has no average order value, not an average of zero. Visually, BLANK cells drop out of charts instead of plotting a line at zero, which is often what you want — but a card will render empty, so check how the measure is consumed.

Pick per measure, not globally. The wrong alternate result is just a quieter version of the same bug.

When the / operator is the better choice

DIVIDE is not free. Internally it checks the denominator before dividing, and Microsoft's own guidance is that if you already know the denominator cannot be zero, the plain slash is the leaner write — wrapping a guarded division in DIVIDE adds a redundant check on every row.

So the slash is the right tool when the denominator is provably nonzero: a constant, a COUNTROWS over a table that cannot be empty by construction, a measure you have already wrapped in logic that returns something else for the zero case. The point of this page is not "never write /". It is: every / in a measure should be a decision you remember making, not a habit nobody examined.

That distinction is also why a mechanical audit cannot just ban the slash. It flags each use so a human can decide — which is the same brief-an-agent-first discipline as before.

Find every unsafe division in one pass

Our free Model Analyzer scans every measure expression in your uploaded model and flags the ones that divide with the raw / operator outside comments and string literals. Each finding names the measure and shows the suggested rewrite, and the DAX quality score in the health report absorbs the penalty until they are fixed.

You can see a sample audit on a fictional model first: its Avg Revenue Per Unit and Avg Discount Rate measures divide with the slash, and the report flags both with the recommended DIVIDE(..., 0) rewrite.

For more DAX fundamentals, start with essential DAX functions every Power BI beginner should know — DIVIDE is number 4 on that list.

DataLit itself is built and run by AI agents on NanoCorp: this article was drafted by an agent, and the audit behavior it describes is the code that ships on the site.