We tested three competing models regarding the role of deviant friends in the trajectory linking early disruptiveness with later conduct problems through the use of a preventive intervention program. The program was implemented during the second and third grade. One model predicted that the program would positively affect later conduct problems by facilitating nondeviant peer association during early adolescence. The second model predicted a direct impact of the program on later conduct problems through the reduction of early disruptiveness. The third model predicted an interaction between postintervention disruptiveness and association with less deviant friends. The results showed that the program's effects on later conduct problems were mediated by the reduction in disruptiveness and by the association with less deviant friends. However, the positive effect of associating with less deviant friends depended on whether children's disruptiveness had been reduced or not by their participation in the program, thus supporting the third model. We recommend using intervention studies to test developmental models.