This dissertation examines the impact of stimulus and response similarity on response selection. Traditional models of response selection invoke a central processor that operates like a look-up table by matching the perceptually classified stimulus (e.g., green square) to the specified response (e.g., right button press). The look-up property of response selection affords the system the ability to map any stimulus onto any response, even if that stimulus-response has never been paired before. Under such an approach, the degree of perceptual similarity or dissimilarity that exists among stimuli in the environment should have little effect on central operations, the similarity or dissimilarity of the motor response executed in response to a stimulus should not influence response selection, and no interaction between stimulus and response features is permitted, given that stimulus features affect the encoding process, and response features affect the output process, but not response selection itself. Eight studies examine the influence of stimulus and response similarity during response selection. The first two experiments establish the interaction across different task demands between stimulus and response similarity. The interaction was not the result of perceptual difficulty (Experiment 3) and was extended to a new set of stimuli (Experiment 4). A consequence of the design in Experiments 1 - 4 was that response condition was confounded with response configuration. In one of the response conditions the target location had three competitors on one side of it compared to the other condition where the target had one competitor on one side and two others on the other side. Experiments 5 and 6 examined the separate roles that response configuration and response metrics had on the interaction between stimulus and response similarity. The mechanism that produced the interaction was the result of competition between partially activated stimulus-response alternatives. Experiments 7 and 8 further explored the role of competition during response selection by turning to traditional response selection methodologies that introduce competition through either the presentation of irrelevant stimulus information or through presenting the stimulus along an irrelevant spatial dimension. These data have broad implications for models of RS. To account for the ability to pair any stimulus modality with any response modality dominant accounts of RS assume that central operations are performed by a generic set of processes that operate over representations that are stripped of metric information (amodal representations). Response selection works as a look-up table that receives a categorized stimulus as an input and returns an abstract response code as output. This type of model cannot produce an interaction between stimulus and response similarity and thus, the present data provide a serious challenge to these types of models. Finally, the data provide evidence that the metric relationship between stimuli and response matter and influence response selection. The co-activation of stimulus-response alternatives are at a level of representation that includes both stimulus and response properties. A framework is presented that captures key aspects of the data.