With its roots in Ungerleider and Mishkin's (1982) uncovering of two distinct — ventral and dorsal — anatomical pathways for the processing of visual information, and boosted by Goodale and Milner's (1992; Milner and Goodale, 1995) behavioral study of patients with lesions of either of these pathways, the perception–action dissociation became a standard reference in the sensorimotor literature. Here we present briefly the anatomical, neuropsychological and, more extensively, the psychophysical evidence favoring such dissociation and pit it against counteracting evidence as well as against potential methodological and conceptual pitfalls. We also discuss classes of models accounting for a number of 'dissociation' results and conclude that the most general and parsimonious one posits the existence of one single processing stream that accumulates information up to a decision criterion modulated by stimulation conditions, response mode (motor vs. verbal/perceptual), task constraints (speeded vs. free time responses) and the nature of the task (detection, discrimination, temporal order judgment, etc.). The reviewed evidence is not meant to refute or validate the hypothesis of a perceptual–motor dissociation. Rather, its main objective is to show that, beyond its self-evidence, such dissociation is difficult if not impossible to test.