Segregation analysis using a regressive model with age- and sex-dependent effects was applied to family data of weight, fat mass (FM) and fat-free mass (FFM) to investigate the major gene hypothesis. The sample consisted of 220 nuclear families from the 'Stanislas Cohort' who volunteered for a free health examination (n = 913). FM and FFM were assessed by bioelectrical impedance. The data were adjusted for height2 and height prior to analysis. The spouse, father-offspring, mother-offspring and sib-sib correlations were: 0.16, 0.18, 0.25 and 0.32 for weight; 0.13, 0.20, 0.23 and 0.28 for FM; 0.18, 0.16, 0.29 and 0.41 for FFM. For the three phenotypes, models specifying a major gene with age- and sex-dependent effects and residual family correlations was better supported than models including only family correlations. For weight, the most parsimonious genetic model was a codominant model with a sex-dependent effect in parents and an age-increasing effect in offspring. For FM, the most parsimonious model was also a codominant model with sex-dependent effects in parents indicating higher effects in women than in men. For FFM, the most parsimonious model was a recessive model with no significant age or sex interaction, although the age interactions paralleled those observed on weight in offspring. For weight and FM, mendelian transmission was rejected. For FFM, the Mendelian and the environmental hypotheses were nearly equally supported and none was rejected when compared to general transmission. Then, evidence for a single major gene could not be inferred for any of the traits. This does not preclude the existence of several genes acting in a more complex way. However, our findings emphasize that weight is a composite phenotype reflecting different components which evolve in distinct ways during life span. For this reason, FM should be highly preferred to weight or BMI for the research of susceptibility genes to obesity.