In this paper we assess the presence of assortative mating, gene-environment interaction and the heritability of intelligence in childhood using a twin family design with twins, their siblings and parents from 112 families. We evaluate two competing hypotheses about the cause of assortative mating in intelligence: social homogamy and phenotypic assortment, and their implications for the heritability estimate of intelligence. The Raven Progressive Matrices test was used to assess general intelligence (IQ) and a persons IQ was estimated using a Rasch model. There was a substantial correlation between spouses for IQ (r = 0.33) and resemblance in identical twins was higher than in first-degree relatives (parents and offspring, fraternal twins and siblings). A model assuming phenotypic assortment fitted the data better than a model assuming social homogamy. The main influence on IQ variation was genetic. Controlled for scale unreliability, additive genetic effects accounted for 67% of the population variance. There was no evidence for cultural transmission between generations. The results suggested that an additional 9% of observed IQ test variation was due to gene-environment interaction, with environment being more important in children with a genetic predisposition for low intelligence.