20Polygenic scores are a popular tool for prediction of complex traits. However, prediction 21 estimates in samples of unrelated participants can include effects of population 22 stratification, assortative mating and environmentally mediated parental genetic effects, a 23 form of genotype-environment correlation (rGE). Comparing genome-wide polygenic score 24 (GPS) predictions in unrelated individuals with predictions between siblings in a within-25 family design is a powerful approach to identify these different sources of prediction. Here, 26we compared within-to between-family GPS predictions of eight life outcomes 27 (anthropometric, cognitive, personality and health) for eight corresponding GPSs. The 28 outcomes were assessed in up to 2,366 dizygotic (DZ) twin pairs from the Twins Early 29Development Study from age 12 to age 21. To account for family clustering, we used mixed-30 effects modelling, simultaneously estimating within-and between-family effects for target-31 and cross-trait GPS prediction of the outcomes. There were three main findings: (1) DZ twin 32 GPS differences predicted DZ differences in height, BMI, intelligence, educational 33 achievement and ADHD symptoms; (2) target and cross-trait analyses indicated that GPS 34 prediction estimates for cognitive traits (intelligence and educational achievement) were on 35 average 60% greater between families than within families, but this was not the case for 36 non-cognitive traits; and (3) this within-and between-family difference for cognitive traits 37 disappeared after controlling for family socio-economic status (SES), suggesting that SES is a 38 source of between-family prediction through rGE mechanisms. These results provide novel 39 insights into the patterns by which rGE contributes to GPS prediction, while ruling out 40 confounding due to population stratification and assortative mating. 41 42 associated with traits other than educational achievement, including intelligence 2,6,7 , 65 socioeconomic status (SES) 8-11 , behaviour problems 12 , mental illness 13 , physical health 13 66 and personality 14,15 , in some cases accounting for as much as or more than the variance in 67 cross-trait associations explained by the target GPS themselves 15,16 . 68 69 However, GWA analyses, and the GPSs derived from them in independent samples, are 70naïve to the pathways that lead from SNPs to trait outcomes 17 . With a focus on prediction, 71 the mechanisms by which polygenic scores relate to phenotypes are left largely unexplored. 72Given the popularity and widespread use of the GPS approach, the interpretation of GPS 73