This article presents an analysis of sample loss in a 4-year follow-up of children aged 4 to 12 who participated in the Ontario Child Health Study in 1983. Of the 1,617 children participating in the original Ontario Child Health Study, 1,172 (72.5%) were located and enlisted at follow-up in 1987. Based on wave-one assessments, nonparticipants at follow-up tended to have higher levels of psychopathology and family risk variables. Respondents were matched with nonparticipants and differentially weighted to compensate for selective loss. In comparing estimates based on actual (observed) and weighted responses in the follow-up sample, it was found that the effects of sample loss depended on the analytical focus. Evaluations of outcome of disorder and risk for disorder were not affected by sample loss. Evaluation of variables that predict persistence of disorder (prognosis) was affected by a bias toward the null.