Prediction from polygenic scores may be confounded sources of passive gene-environment correlation (rGE; e.g. population stratification, assortative mating, and environmentally mediated effects of parental genotype on child phenotype). Using genomic data from 10,000 twin pairs, we asked whether polygenic scores from the recent externalising genome-wide association study predicted conduct problems, ADHD symptomology and callous-unemotional traits, and whether these predictions are biased by rGE. We ran regression models including within-family and between-family polygenic scores, to separate the direct genetic influence on a trait from environmental influences that correlate with genes (indirect genetic effects). Findings suggested that this externalising polygenic score is a good index of direct genetic influence on conduct and ADHD-related symptoms across development, with minimal bias from rGE, although the polygenic score predicted less variance in CU traits. Post-hoc analyses showed some indirect genetic effects acting on a common factor indexing stability of conduct problems across time and contexts.