Repetition priming is often thought to reflect the facilitation of 1 or more processes engaged during initial and subsequent presentations of a stimulus. Priming can also reflect the formation of direct, stimulus-response (S-R) bindings, retrieval of which bypasses many of the processes engaged during the initial presentation. Using long-lag repetition priming of semantic classification of visual stimuli, the authors used task switches between study and test phases to reveal several signatures of S-R learning in Experiments 1 through 5. Indeed, the authors found surprisingly little, if any, evidence of priming that could not be attributed to S-R learning, once they considered the possibility that stimuli are simultaneously bound to multiple, different response codes. Experiments 6 and 7 provided more direct evidence for independent contributions from at least 3 levels of response representation: the action (e.g., specific finger used), the decision (e.g., yes-no), and the task-specific classification (e.g., bigger-smaller). Although S-R learning has been discussed previously in many contexts, the present results go beyond existing theories of S-R learning. Moreover, its dominant role brings into question many interpretations of priming during speeded classification tasks in terms of perceptual-conceptual processing. (Contains 4 figures, 5 tables and 6 footnotes.)