The impact of non-response (NR) in baseline data collection on the ability to analyze treatment effects in Project ALERT (Adolescent Experiences in Resistance Training) was studied. Project ALERT is a multisite, multiyear test of a smoking and drug prevention program for seventh and eighth graders. Questionnaires about drug use and related topics were administered to students enrolled in 20 treatment and 10 control schools in 8 California and Oregon districts, and saliva samples were taken. By April 1988, six data collection waves were completed. The baseline population was 7,566 students. Of the intended sample, 16% did not participate in some or all of the baseline data collection because of parents' refusal to allow participation, students' refusal to participate, and absence from school on survey dates. In all, 1,484 respondents and 1,287 non-respondents were selected for analyses comparing the two groups in terms of grades, absenteeism, race/ethnicity, and gender. The number of students lost due to baseline non-response (NR), differences between respondents and non-respondents, the effect of NR on the sample, and potential differences among treatment groups caused by NR were also analyzed. The likelihood that a high proportion of non-responders were substance users or at high risk of using substances and the threat to internal validity caused by NR were important parts of the analyses. The results show that: non-respondents and respondents differed substantially; NR had little impact on the sample; internal validity was maintained; and steps to reduce NR succeeded. Six data tables and two bar graphs are included. (TJH)