Associations between nominated best friends’ behavior at T1 – including those who nominated no best friends at T1 as an additional comparison group – and target child's behavior at T1 were investigated using multivariate analysis of variance (MANOVA), with the target child's antisocial and prosocial scores entered as multivariate dependent variables. Longitudinal associations between nominated best friends’ behavior at T1 and changes in target child's antisocial and prosocial behavior from T1 to T2 were also investigated using MANOVA. To this end, the variance in T2 target child behavior scores explained by T1 target child scores for the same behavior was first partialed out by saving the unstandardized residuals from a regression of T2 behavior on T1 behavior. These residual values were then entered as the T2 dependent variables in the subsequent MANOVA. In both the concurrent and longitudinal analyses, nominated best friends’ behavior was entered as a 5-level between-subjects independent variable (no friends, moderate antisocial/high prosocial, moderate antisocial/low prosocial, low antisocial/high prosocial, and low antisocial/low prosocial). Gender and age were also included as between-subjects independent variables in both equations, with age dichotomized into a younger (n = 97) versus older (n = 106) group by dividing T1 age in months at the median (51 months). A customized model including main effects and all two-way interactions was tested.