WGJ collected data from 7,583 students attending 11 middle schools in Philadelphia. Their analysis revealed that community-level differences between schools added only slightly to the explanatory power provided by individual-level constructs. The present research used a much larger sample of schools. The data came from the eighth-grade panel of the National Educational Longitudinal Study, a national probability sample collected in the fall of 1988. The 17,424 adolescents from 1,014 schools responded to a baseline personal interview at the end of the first semester of the 1988-89 school year and a follow-up personal interview in 1990. The model included students as level-one units and schools as level-two units. Results revealed that the effects of all eight student-level explanatory variables, as well as the school intercept, varied substantially depending on the school. Findings suggested that schools have a much more important role in conditioning the associations between individual explanatory variables and misconduct than in directly determining misconduct. As in the WGJ analysis, the direct effects of school variables were small and were statistically insignificant, except for the effect of teacher salary. Findings demonstrated the power of multilevel models to test theoretical propositions about macro-micro links and to determine whether relationships at the individual level vary at the school or community level. Tables, footnotes, and 29 references (Author abstract modified)