The empirical study of legislative behavior largely relies on roll-call-vote analysis, but roll-call votes in many legislatures represent only a sample of legislative votes. We have good reasons to believe this sample is particularly poor for inferring party effects on legislative behavior. The selection of votes for roll call may be endogenous to exactly the characteristics of voting behavior (for instance, party cohesion) that we want to study. We must understand the roll-call-vote institution and account for its selection effects before we can draw inferences about legislative behavior from roll-call results. This paper develops a game-theoretic model of roll- call-vote requests predicated on party leaders requesting votes to enforce party discipline. The model offers general and testable predictions about the selection process and how it affects observed and unobserved legislative voting behavior, particularly party cohesion.