This paper contributes empirically to the literature on peer effects in first-grade classrooms. The paper examines peer effects on academic achievement among first graders randomly assigned to their classrooms and to their teachers as part of Tennessee's Project STAR, America's largest ever education experiment. The analysis draws on previously unexploited measures of kindergarten achievement taken before random assignment to first-grade classes and available for about sixty percent of this sample. Data are not missing at random. This paper studies effects of peer lagged achievement on first-grade achievement. The STAR data allow for credible inference about peer effects because students and teachers were randomly assigned to classes within school. Further, the data contain pre-assignment measures of achievement, which are useful as conditioning variables to explain each student's own outcome and for characterizing each student's peer group. The paper contributes methodologically to the larger peer-effects literature in advancing the understanding of how to make inference about peer effects in the presence of missing data on peers.