Peer similarity in delinquency has been studied extensively. But basic questions remain about measuring peer delinquency and how important the nature of relationships with delinquent peers is. This article uses data from the NSCR School Project, which has collected unusually detailed information about delinquent peers and the social networks of adolescents. We examine differences in the roles of regular friends and best friends with regard to peer similarity in delinquent behavior. We also contrast two methods of measuring peer delinquency: the conventional one of asking respondents about their peers, and the social network method, by which peers report about themselves. The results show that respondents can have best and regular friends who differ in their degree of delinquency, and that the association between respondent and peer delinquency does not differ much between friends and best friends. At the same time, our results suggest that both types of peers influence the level of respondent delinquency. Measures based on the direct network method resulted in higher estimates of peer delinquency, but in lower estimates of the association between respondent and peer delinquency.