Modeling individual fish habitat selection in highly variable environments such as hydropeaking rivers is required for guiding efficient management decisions. We analyzed fish microhabitat selection in the heterogeneous hydraulic and thermal conditions (modelled in two-dimensions) of a reach of the large hydropeaking Rhône River locally warmed by the cooling system of a nuclear power plant. We used modern fixed acoustic telemetry techniques to survey 18 fish individuals (five barbels, six catfishes, seven chubs) signaling their position every 3 s over a three-month period. Fish habitat selection depended on combinations of current microhabitat hydraulics (e.g. velocity, depth), past microhabitat hydraulics (e.g. dewatering risk or maximum velocities during the past 15 days) and to a lesser extent substrate and temperature. Mixed-effects habitat selection models indicated that individual effects were often stronger than specific effects. In the Rhône, fish individuals appear to memorize spatial and temporal environmental changes and to adopt a "least constraining" habitat selection. Avoiding fast-flowing midstream habitats, fish generally live along the banks in areas where the dewatering risk is high. When discharge decreases, however, they select higher velocities but avoid both dewatering areas and very fast-flowing midstream habitats. Although consistent with the available knowledge on static fish habitat selection, our quantitative results demonstrate temporal variations in habitat selection, depending on individual behavior and environmental history. Their generality could be further tested using comparative experiments in different environmental configurations.