The first wave of data collection took place in 1988 and used a systematic probability sample of 1,208 students in 9th, 10th, and 11th grades in 3 community high school. Follow-up interviews took place in 1989, 1990, 1993, and 1996. The final interviews included 92 percent of those in the previous wave and 69 percent of the original sample. The predictors of drinking included stressful life events, social support from parents and peers, and relationship problems with parents and peers. The data analysis used HLM, a statistical program for the analysis of multilevel or hierarchical data, and estimated a two-level hierarchical linear model to model individual trajectories in drinking behavior. Results indicated that stressful life events and parental support and conflict had significant effects on both the frequency and the intensity of alcohol use. Age-related changes in these variables coincided with changes in drinking behavior, but they did not account for the variability of drinking over this period. Results from conditional models demonstrated that the impact of the stress was contingent on age and that the strong associations between drinking and stress during the high school years weakened considerably as individuals moved into their late teens and early twenties. Figures and 40 references (Author abstract modified)