Relations between adolescents' substance use and perceptions of their friends' substance use were examined cross-sectionally and longitudinally in a predominantly African-American school district. Fourth- and fifth-grade students were surveyed and tracked for 4 consecutive years. Cross-sectional samples included 3,073, 5,955, 7,701, and 6,616 students in years 1 to 4, respectively; the longitudinal sample included 1,802 students surveyed in every year. Self-reported substance use of friends and classmates also was assessed. Perceived friends' substance use had a stronger association with prior substance use than friends' self-reported substance use in every year. Perceived family use and classmates' self-reported use also made independent contributions to regression models. Longitudinal structural equation analyses indicated that perceived friends' use is more likely to be a product of an adolescent's previous substance use than a precursor of subsequent substance use. The findings contradict prevailing theories on the influence of peers on substance use.