In populations of phytophagous insects that use the host plant as a rendezvous for mating, divergence in host preference could lead to sympatric speciation. Speciation requires the elimination of “generalist” genotypes, that is, those with intermediate host preference. This could occur because such genotypes have an inherent fitness disadvantage, or because preference alleles become associated with alleles that are oppositely selected on the two hosts. Although the former mechanism has been shown to be plausible, the latter mechanism has not been studied in detail. I consider a multilocus model (the “Bush model”) in which one set of biallelic loci affects host preference, and a second set affects viability on the hosts once chosen. Alleles that increase viability on one host decrease viability on the other, and all loci are assumed to be unlinked. With moderately strong selection on the viability loci, preference alleles rapidly become associated with viability alleles, and the population splits into two reproductively isolated host specialist populations. The conditions for speciation to occur in this model, as measured by the strength of selection required, are somewhat more stringent than in a model in which preference and viability are controlled by the same loci (one-trait model). In contrast, the conditions are much less stringent than in a model in which speciation requires buildup of associations between viability loci and loci controlling a host-independent assortative mating trait (canonical two-trait model). Moreover, in the one-trait model, and to a lesser extent the Bush model, the strength of selection needed to initiate speciation is only slightly greater than that needed to complete it. This indicates that documenting instances of sympatric species that are reproductively isolated only by host or habitat preference would provide evidence for the plausibility of sympatric speciation in nature.