We use empirical data to demonstrate a recently proposed framework for including measured genotypes in structural equation models (Van den Oord and Snieder, 2002). The sample consisted of 227 MZ pairs, 197 DZ pairs, and 41 DZ twins without co-twin. The mean age of the female twins was 57 years. The analysis aimed at studying the effect of the vitamin D receptor (VDR) gene polymorphism (VDRGP) on bone mineral density (BMD). Whereas age affected the BMD measures via independent pathways, the effects of the VDRGP plus latent genetic and environmental factors were consistent with a common pathway model. VDRGP explained 0.3% to 0.5% of the variance. These analyses demonstrated that (i) regardless of how much is known about the gene and its products at a molecular level, meaningful questions can be addressed concerning how genes co-act with other variables at a "higher" phenotypic level; (ii) measured genotypes can be treated like any other variable so that the whole structural equation models framework can in principle directly be applied; and (iii) measured genotypes do not need to have large effects to study mechanisms underlying complex traits, and it may even be argued that particularly genes with small effects are better studied within the context of models with multiple variables.