We analyzed a population-representative cohort (N = 13,611; Mage at kindergarten, first, and second grade = 67.5, 79.5, and 91.5 months, respectively) to identify kindergarten to second grade factors predictive of being bullies or victims during third to fifth grade. We did so by estimating a block recursive structural equation model (SEM) with three sets of predictors. These were: (a) individual and school socio-demographics; (b) family distress and harsh parenting; and (c) individual behavior and achievement. Relations between each of the included variables and the bullying outcomes were simultaneously estimated within the SEM. Thus, each variable served as a control for estimating the effects of the other variables. We used robust standard errors to account for student clustering within schools. Results indicated that externalizing problem behavior strongly predicted being a bully ([ES] = 0.56, p < 0.001) and a victim (ES = 0.29, p < 0.001). We observed a negative relation between being Hispanic and being a victim (ES = − 0.10, p < 0.001) and a positive relation between being Black and being a bully (ES = 0.11, p < 0.001). We also observed statistically significant relations between a family’s socioeconomic status and being a bully (ES = − 0.08, p < 0.001) as well as school poverty and being a victim (ES = 0.07, p < 0.001). The results advance the field’s limited understanding of risk and protective factors for bullying perpetration or victimization during elementary school and provide additional empirical support for assisting young children already exhibiting externalizing problem behaviors.