In the study of complex diseases, it may be important to test hypotheses related to gene-gene (G x G) interaction. The success of such studies depends critically on obtaining adequate sample sizes. In this paper, the author investigates sample size requirements for studies of G x G interaction, focusing on four study designs: the matched-case-control design, the case-sibling design, the case-parent design, and the case-only design. All four designs provide an estimate of interaction on a multiplicative scale, which is used as a unifying theme in the comparison of sample size requirements. Across a variety of genetic models, the case-only and case-parent designs require fewer sampling units (cases and case-parent trios, respectively) than the case-control (pairs) or case-sibling (pairs) design. For example, the author describes an asthma study of two common recessive genes for which 270 matched case-control pairs would be required to detect a G x G interaction of moderate magnitude with 80% power. By comparison, the same study would require 319 case-sibling pairs but only 146 trios in the case-parent design or 116 cases in the case-only design. A software program that computes sample size for studies of G x G interaction and for studies of gene-environment (G x E) interaction is freely available (http://hydra.usc.edu/gxe).