Edsger W. Dijkstra, 1969

AI-generated editorial portrait of Edsger W. Dijkstra in a late-1960s university classroom beside program-logic diagrams

“Program testing can be used to show the presence of bugs, but never to show their absence!”

Edsger W. Dijkstra framed this warning after the 1969 NATO Software Engineering Conference, when the young software field was confronting projects whose complexity outpaced familiar programming habits. Testing could demonstrate that a program failed on a particular input, under a particular condition, at a particular moment. A successful run, however, established much less. It showed only that the tested path had not revealed a failure. For any nontrivial program, the possible combinations of inputs, states, timing, hardware behavior, and external dependencies are too numerous to exhaust. Dijkstra’s sharp contrast turned that asymmetry into a lesson engineers could remember.

The quote is sometimes misread as an argument against testing. Dijkstra’s point was almost the opposite: testing is valuable evidence, but its evidence has limits. A failing test is decisive because one counterexample disproves a universal claim that the program always behaves correctly. A passing test is conditional because countless untested cases remain. The distinction resembles mathematical reasoning. Examples can help uncover a false conjecture, while a proof must account for every case covered by the claim. Software engineering cannot always supply a complete formal proof, but specifications, invariants, type systems, code review, and structured design can make the reasoning broader than a collection of examples.

Contemporary development combines those forms of assurance. Unit tests check small behaviors, integration tests examine interactions, and end-to-end tests exercise complete user journeys. Property-based testing explores many generated cases, fuzzing searches for malformed inputs, static analysis identifies suspicious patterns, and formal methods can verify selected safety properties. None is a universal guarantee. Their strength comes from overlap: each method observes a different class of risk and exposes assumptions the others may miss. A team that treats a green test suite as proof of perfection can overlook race conditions, security failures, rare states, and misunderstandings in the requirements themselves.

Dijkstra’s warning is especially relevant when software acts at scale. A defect in a local tool may inconvenience one user; the same defect in a payment network, medical device, or cloud service can spread rapidly. Confidence therefore should be stated in terms of evidence: what was tested, what properties were checked, which environments were covered, and what remains unknown. The goal is not to eliminate uncertainty through a slogan, but to manage it honestly. Testing gives engineers indispensable ways to find failures and prevent regressions. Dijkstra asks them to pair that practice with disciplined reasoning, because absence of observed failure is not the same thing as demonstrated correctness.

Dijkstra used the line while discussing the intellectual challenge of programming after the 1969 NATO software engineering meeting in Rome.

He was responding to the difficulty of establishing confidence in increasingly complex programs, not dismissing practical testing work.

One failing test is a counterexample, while any finite collection of passing tests leaves unexamined cases.

Testing becomes stronger when paired with precise specifications, structured design, invariants, reviews, and appropriate formal analysis.

Modern systems add concurrency, networks, dependencies, and adversarial inputs, multiplying the states a test suite cannot fully enumerate.

Responsible teams describe the evidence behind confidence instead of treating a green dashboard as proof that no defect exists.

Explore more "Quotes of The Day"

Discover more notable quotes from influential voices across politics, science, business, technology, sports, and culture. Each quote offers insight into how ideas, beliefs, and decisions shape the world around us.