Over the last decade, implicit measures of mental associations (e.g., Implicit Association Test, sequential priming) have become increasingly popular in many areas of psychological research. Even though successful applications provide preliminary support for the validity of these measures, their underlying mechanisms are still controversial. The present article addresses the role of a particular mechanism that is hypothesized to mediate the influence of activated associations on task performance in many implicit measures: response interference (RI). Based on a review of relevant evidence, we argue that RI effects in implicit measures depend on participants attention to association-relevant stimulus features, which in turn can influence the reliability and the construct validity of these measures. Drawing on a moderated-mediation model (MMM) of task performance in RI paradigms, we provide several suggestions on how to address these problems in research using implicit measures.