The sampling variances in twin and sibling studies of man for estimation of intraclass correlation coefficients and their relation to heritabilities are evaluated. For twins and full-sibs the higher the estimated heritability of a trait the smaller the sampling variance. With increase in the number of twin sets or of full-sibships studied, the sampling variances decrease by their proportions. Twin and full-sib studies with lower heritabilities (h2 < 0.25) need at least 50 sibships to obtain a significant estimate. Using half-sibs to estimate heritabilities, increase of the offspring number from the same parent would be more efficient for reducing the sampling variance of the heritability compared with increases of half-sibship families. However, because of the limited number of offspring per family in modern man, the efficiency of the half-sib design could be increased by the large number of monozygotic twin families. It appears that with only 2 or 3 children per full-sibship family, approximately 100 half-sibships are needed for significant estimate of heritability.