The aim of this thesis is to show the variety of information which can be obtained from sensitive analysis of data on MZ and DZ twins reared together, the most popular design in human behaviour genetics. Initial theoretical considerations suggest the range of models it is possible to fit to the data. We introduce a method of calculating power based on the non-central chisquare distribution. This allows calculation of sample sizes required to reject false hypotheses with given probability and it is shown that even in some simple situations, very large sample sizes are required for reliable rejection. Our calculations suggest that the power of the three twin studies analysed to discriminate between simple alternative models, is rather low. A twin study of examination performance is able to reject environmental models for variation in most variables and detects large heritabilities for both scholastic achievements and IQ. Significant heterogeneity of gene action is found for English and Mathematics suggesting specific genetical variation for those subjects. A larger study of sexual and social attitudes and personality finds several differences in patterns of variation between the sexes. Many genotype-environment interactions are demonstrated for certain scales of measurement but rescaling the data to remove them makes little difference to the results of genetical analysis in most cases. Evidence for the genetical effects of competition between twin sisters is found for the trait of Sexual Satisfaction and differences between males and females in the relative importance of cultural and genetic variation are found in Libido, Masculinity and Toughmindedness. There is directional, but not significant evidence of higher heritabilities for personality traits in men. Because of variation in the age at which different genotypes begin sexual activity the heritability of age of first sexual intercourse increases with age itself. Surprisingly, there is no evidence of cultural effects influencing this trait. The behaviour of twin boys interacting with their parents was recorded by ethological observation in both twins and singletons. Most of the measurements were too unreliable to hope to detect genetic variation with such small samples, the only exceptions being child speech and a measure of rudimentary conscience. However, some new evidence against arbitrary parent-initiated treatment inequalities for MZ and DZ twins provides further support for the validity of the twin method. A method for the genetical analysis of covariance structures is developed which allows simultaneous maximum likelihood estimates of general and specific variation from the various genetical and environmental sources. The method is illustrated on some twin data for cognitive abilities. Possibilities for future research arising from the results of each of these twin studies are discussed.