The refuge strategy is used widely for delaying evolution of insect resistance to transgenic crops that produce Bacillus thuringiensis (Bt) toxins. Farmers grow refuges of host plants that do not produce Bt toxins to promote survival of susceptible pests. Many modelling studies predict that refuges will delay resistance longest if alleles conferring resistance are rare, most resistant adults mate with susceptible adults, and Bt plants have sufficiently high toxin concentration to kill heterozygous progeny from such matings. In contrast, based on their model of the cotton pest Heliothis virescens, Vacher et al. (Journal of Evolutionary Biology, 16, 2003, 378) concluded that low rather than high toxin doses would delay resistance most effectively. We demonstrate here that their conclusion arises from invalid assumptions about larval concentration-mortality responses and dominance of resistance. Incorporation of bioassay data from H. virescens and another key cotton pest (Pectinophora gossypiella) into a population genetic model shows that toxin concentrations high enough to kill all or nearly all heterozygotes should delay resistance longer than lower concentrations.