Numerous studies have found interactions between socioeconomic status (SES) and the heritability of cognitive ability in samples from the United States, with individuals from lower SES backgrounds showing decreased heritability compared to those reared in higher SES environments. However, nearly all published studies of the Scarr-Rowe interaction have been univariate and cross-sectional. In this study, we sought to maximize statistical power by fitting multivariate models of gene (G) x SES interaction, including longitudinal models. Cognitive ability data collected at up to five time points between ages 7 and 15 years were available for 566 twin pairs from the Louisville Twin Study. We used multilevel and latent factor models to pool intelligence subtest scores cross-sectionally. To examine interactions longitudinally, we fit latent growth curve models to IQ scores. Power analysis results indicated that the multivariate approach substantially boosted power to detect G x SES interaction. The predicted interaction effect was observed at most ages in cross-sectional multivariate analyses. In longitudinal analyses, we found significant G x SES interactions on mean-level (intercept) full scale IQ and performance IQ (ps < .001), but not verbal IQ intercept (p = .08). SES did not significantly moderate the heritability of change in IQ over time (slope). Interaction appeared to be driven by DZ twin correlations decreasing more substantially as a function of higher SES than MZ correlations.