The most definitive evidence about any aspect of stuttering is that listeners are unable to judge unit-by-unit occurrences of it acceptably. This result has been replicated repeatedly in every decade for a half century. Nonetheless, for virtually all research and most clinical practice, stuttering has been defined perceptually as if listeners could identify it accurately. Reasons for this state of affairs and its implications for therapy, theory, and research are analyzed. An alternative speech production definition with its implications is proposed. Further, a diagnostic method of validating authentic stuttering is described, as is an objective for fluency-skill therapy that reduces rather than reinforces avoidance behavior.