Prey refuges are expected to affect population dynamics, but direct experimental tests of this hypothesis are scarce. Larvae of western flower thrips Frankliniella occidentalis use the web produced by spider mites as a refuge from predation by the predatory mite Neoseiulus cucumeris. Thrips incur a cost of using the refuge through reduced food quality within the web due to spider mite herbivory, resulting in a reduction of thrips developmental rate. These individual costs and benefits of refuge use were incorporated in a stage-structured predator–prey model developed for this system. The model predicted higher thrips numbers in presence than in absence of the refuge during the initial phase. A greenhouse experiment was carried out to test this prediction: the dynamics of thrips and their predators was followed on plants damaged by spider mites, either with or without web. Thrips densities in presence of predators were higher on plants with web than on unwebbed plants after 3 weeks. Experimental data fitted model predictions, indicating that individual-level measurements of refuge costs and benefits can be extrapolated to the level of interacting populations. Model-derived calculations of thrips population growth rate enable the estimation of the minimum predator density at which thrips benefit from using the web as a refuge. The model also predicted a minor effect of the refuge on the prey density at equilibrium, indicating that the effect of refuges on population dynamics hinges on the temporal scale considered.