What is pairwise testing and why is it useful?

Pairwise (or all-pairs) testing rests on a result from combinatorial testing research: most defects in systems with multiple inputs depend on the interaction of just two inputs at a time. By covering every pair of input combinations, you catch the vast majority of those bugs without enumerating every full combination.

The math is dramatic. A feature with 5 inputs, each with 4 values, has 4^5 = 1024 full combinations. The minimum number of test cases needed to cover every pair is around 16 — a 60× reduction.

When pairwise is the right tool:

• Many inputs, many values per input (full combinatorial is impractical).
• Inputs combine in unknown ways (you don't have a decision table).
• Risk profile is "any pair could be buggy" — typical for configuration UIs, search filters, form combinations.

How to apply it:

1. List inputs and their possible values.
2. Use a tool — PICT (Microsoft), AllPairs.py, Hexawise — to generate the minimum set covering every pair.
3. Run the generated cases. If a defect is found involving three or more inputs, add explicit cases (pairwise won't catch those).

Limits worth knowing:

• Three-way bugs slip through. If a defect requires admin + UK + free together but not any pair, pairwise misses it. Higher-order (t-way) techniques exist but cost more cases.
• Pairwise doesn't tell you the right answer. It tells you which combinations to test; you still have to know the expected behaviour for each.

NIST research suggests ~70% of defects involve 1 or 2 factors; ~95% involve ≤ 4. Pairwise hits the sweet spot for most projects.