Genetic association studies are becoming commonplace due to the availability of cost-effective yet sophisticated DNA sequencing and genotyping resources and technologies. In addition, technologies designed to identify molecular and subclinical phenotypes that reflect disease pathogenesis are continually being developed and refined (consider, e.g., imaging technologies, microarray-based gene expression and proteomic platforms, histological analyses of excised tissues, etc.). Unfortunately, the large-scale use of many of these molecular and subclinical phenotyping technologies in genetic association studies is difficult logistically and is currently cost-prohibitive. In this paper, we consider efficient designs for testing the association between particular genetic variations and expensive, yet appropriate, subclinical phenotypes of relevance to a disease that take advantage of twins or sibling pairs discordant for genotypes at the locus (or loci) being tested. We demonstrate that including genotypically discordant twins or siblings in an association study can result in a substantial increase in power over designs that use monozygotic twins or only unrelated individuals. We ultimately argue that, from a practical standpoint, sampling from existing family or twin-based cohorts in which: (1) follow-up studies of a genetic association are warranted in order to assess the in vivo significance of an association with respect to more refined pathological phenotypes; and/or (2) large-scale, genome-wide linkage and association studies have been pursued that have focused on clinical endpoints for which the study subjects have consented to more elaborate follow-up studies, is a powerful way to test associations.