Longitudinal studies conducted with children typically experience non-response problems that can make substantial inroads on sample size and introduce bias into the analysis. Three methods for limiting non-response in school-based research programs were assessed: (1) asking parents to return consent forms only if they do not want their children to participate in the research; (2) conducting make-up sessions for students who are absent during the scheduled data collection day; and (3) tracking students who transfer out of the participating schools. These methods were used in a large-scale intervention study, involving nearly 10,000 junior high school students, designed to assess the effects of a smoking and drug use prevention curriculum for adolescents (Project ALERT) across 30 schools in 8 school districts in California and Oregon. The evaluation data, which consist of surveys and saliva samples, were collected in the classrooms before the 7th grade curriculum was delivered and 3, 12, 24, and 36 months after it was delivered. Each method substantially reduced the anticipated loss of participating students over a 2-year period; each method also reduced attrition bias attributable to the loss of at-risk students (those with characteristics such as poor grades, prior deviance, and past drug experience that made them likely candidates for using drugs in the future). By using a multistage parent consent process that included multiple communication channels, the risk of mistakenly treating non-response as consent was minimized, and the loss of between 40% and 50% of the targeted population at the onset was avoided. Over several waves of data collection, make-up sessions reduced the losses attributable to absenteeism by an average of about 60%, recapturing close to 6% of the baseline sample each year. The tracking process retrieved surveys from two-thirds of the transferees, and cut the overall attrition rate in half. Two tables and one figure illustrate the study. (SLD)