Setting‐level norms are typically measured by using the mean of individual reports. Investigating norms for nonviolent problem solving, this study assessed the added value of two additional characteristics: the range of acceptable behaviors and the degree of consensus among individuals in each setting. Using data from 5,386 participants in a violence prevention project (74 classes), this study estimated the cross‐sectional and longitudinal effects of norm characteristics on aggression, beliefs about aggression, self‐efficacy for nonviolence, and beliefs about nonviolence. Each norm measure showed acceptable setting‐level reliability. Class mean approval predicted beliefs about aggression, self‐efficacy, and beliefs about nonviolence. With class means already in the model, the range of acceptable behaviors predicted additional cross‐sectional variance in all four outcomes. Consensus predicted aggression and beliefs about nonviolence beyond other characteristics. Mean approval and consensus had effects on growth in at least one outcome. Discussion focuses on the importance of setting‐level measurement and implications for intervention. © 2010 Wiley Periodicals, Inc.