Stimulus-response (S-R) episodes are formed whenever a response is executed in close temporal proximity to a stimulus. Subsequent stimulus repetition will retrieve the episode from memory, reactivating the previous response. Whereas many research findings attest to the flexibility of representing stimulus features, only little is known about the way responses are coded within transient S-R episodes, that is, whether the retrieved response is represented in terms of specific motor codes, abstract/semantic codes, or both. To differentiate between these accounts, we used an approach/avoidance task in which semantic meanings (i.e., moving a word “towards” or “away from” a manikin on screen) and motor codes of responses (i.e., pulling or pushing a joystick) were manipulated orthogonally. Results of 2 experiments indicated that stimulus repetitions retrieve both, semantic as well as motor code representations, indicating multiple and independent levels of response coding. We conclude that response representation in S-R episodes follows similar binding principles as are known from stimulus integration.