In recent years, there has been a focus on to understand the etiology of complex diseases which do not follow simple Mendelian, single-locus segregation. Complex diseases are assumed to be the result of more than one locus and/ or environmental factors and are thought to be influenced by many intermediate quantitative traits. Quantitative traits have been studied extensively by plant and animal geneticists. With the advent of new tools and methods, comprehensive approaches to identify candidate genes underlying quantitative traits for humans are available. Testing the contribution of candidate genes to quantitative trait variation will become commonplace as more and more genes are identified. One of the methods that shows great promise in the analysis of candidate genes involves "Variance Components" models originally introduced by Fisher. Fisher partitioned the total genetic variance into additive part resulting from additive effects or main effects of the genes, a part resulting from dominant effects (allelic interactions) of genes, and a part resulting from epistatic effects (non-allelic interactions) of genes. Cockerham further partitioned epistatic variance into additive x additive, additive x dominant, dominant x dominant components of variance. Tiwari and Elston have shown that epistatic components of variance explain the significant part of the total genetic variance in certain two locus models pertinent to human genetics models. In this paper, we investigate the sample size required to achieve pre-specified power to detect epistatic components of variance for the twolocus models. This investigation includes analytical as well as simulation techniques for the development of the proposed method and illustration through examples.