Animals commonly choose among habitats that differ both in foraging return and mortality hazard. However, no experimental study has attempted to predict the level of increase in resources, or the decrease in mortality hazard, which will induce a forager to shift from a safer to a more hazardous (but richer) foraging area. Here we present and test a model that specifies the choice of foraging areas ("habitats") that would minimize total mortality risk while allowing collection of some arbitrary net energy gain. We tested the model with juvenile creek chubs (Semotilus atromaculatus) in an experimental field stream in which the foragers could utilize a foodless refuge and choose between two foraging areas that differed in experimentally manipulated resource densities (Tubifex spp. worms in sediments) and mortality hazard (adult creek chubs). For the case tested, the model specified a simple rule: "use the refuge plus the site with the lowest ratio of mortality rate (µ) to gross foraging rat (f)," i.e., "minimize µ./f." Independent prior measurements of mortality hazard (as a function of predator density) and gross foraging rate (as a function of resource density) allowed us to predict the resource level in the more hazardous foraging site that should induce a shift from the safer to the more hazardous site. The chubs' preferences in subsequent choice experiments agreed well with the theoretical predictions. The "minimize µ/f" rule (deaths per unit energy), perhaps in modified form, provides a simple alternative to the "maximize f" (energy per unit time) criterion that applies to long-term rate maximization when predation hazard does not differ among choices.