This paper discusses linkage detection strategies for genetically complex traits using multilocus models. Two types of models are described: a multiplicative model representing epistasis (gene-gene interaction) and an additive model approximating genetic heterogeneity. The risk ratio λR, comparing the risk of a relative to population prevalence, is defined. For a single-locus model, λR - 1 decreases by a factor of two with each degree of relationship. For an additive model, the same holds true. For a multiplicative model, λR - 1 decreases more rapidly. Examination of λR values for various relatives can suggest multiple loci and epistasis. For example, data for schizophrenia suggest multiple loci in interaction. The second paper in this series shows that λR is the critical parameter in determining power to detect linkage using affected relative pairs. The paper examines single- and multiple-locus models. For a single-locus model, λR - 1 decreases by a factor of two with each degree of relationship. If dominance variance is zero, the same holds for MZ twins and sibs. For a two-locus multiplicative model, λR is the product of risk ratio factors for each locus. For an additive model, λR - 1 is a weighted sum of similar terms for each locus. For a genetic heterogeneity model, λR - 1 is a combination of terms from each locus. The paper also extends these models to additional loci and discusses the implications for schizophrenia. The example of schizophrenia shows that multiple loci and epistasis are likely involved. The paper concludes that significant epistatic variance components are necessary to explain the observed risk patterns. The results suggest that multiple loci are involved in schizophrenia, and that the risk ratios decrease with increasing degree of relationship. The paper emphasizes the importance of λR in determining linkage detection power.This paper discusses linkage detection strategies for genetically complex traits using multilocus models. Two types of models are described: a multiplicative model representing epistasis (gene-gene interaction) and an additive model approximating genetic heterogeneity. The risk ratio λR, comparing the risk of a relative to population prevalence, is defined. For a single-locus model, λR - 1 decreases by a factor of two with each degree of relationship. For an additive model, the same holds true. For a multiplicative model, λR - 1 decreases more rapidly. Examination of λR values for various relatives can suggest multiple loci and epistasis. For example, data for schizophrenia suggest multiple loci in interaction. The second paper in this series shows that λR is the critical parameter in determining power to detect linkage using affected relative pairs. The paper examines single- and multiple-locus models. For a single-locus model, λR - 1 decreases by a factor of two with each degree of relationship. If dominance variance is zero, the same holds for MZ twins and sibs. For a two-locus multiplicative model, λR is the product of risk ratio factors for each locus. For an additive model, λR - 1 is a weighted sum of similar terms for each locus. For a genetic heterogeneity model, λR - 1 is a combination of terms from each locus. The paper also extends these models to additional loci and discusses the implications for schizophrenia. The example of schizophrenia shows that multiple loci and epistasis are likely involved. The paper concludes that significant epistatic variance components are necessary to explain the observed risk patterns. The results suggest that multiple loci are involved in schizophrenia, and that the risk ratios decrease with increasing degree of relationship. The paper emphasizes the importance of λR in determining linkage detection power.