In the framework of a research project focusing on mitigation measures for hydropeaking, a lateral embayment in riverbank is studied as a fish shelter. With the goal to find attractive configurations of shelter, systematic experiments with wild juvenile brown trout (Salmo trutta fario) were carried out in a flume supplied with freshwater from a natural river. The experimental is equipped with a rectangular lateral embayment. In order to follow the fish trajectories, their movements were recorded continuously by video camera during every test, and their positions were periodically observed. In order to link the swimming trajectories of the trout with the flow conditions, 2D simulations were computed to obtain the diverted discharge in the shelter, and systematic UVP measurement of the velocity field was performed. The flow velocities were analysed in the vertical interface between the shelter and the flume. Comparing the velocity patterns with the fish trajectories, the attractiveness of different configurations of fish shelters could be analyzed. The first tests reveal that a very basic shelter configuration, with low water exchange between shelter and channel, is not interesting for fish. When forcing a water exchange By introducing a deviation groyne into the shelter with the aim to force the water exchange, the frequentation rate can be increased significantly. The fish can easily detect the refuge by the exchange flux when searching its way upstream. The shelter attractiveness was optimized by testing different groyne orientations, in order to create an expanded velocity field close to the exit and entrance sections. Important is a high velocity field leaving the refuge at its lower end but also a backwater zone near the groyne. The high velocity field attracts the fish and the close backwater zone allows the fish to enter the shelter. For the best configuration, more than 80% of the fish found the shelter by swimming mainly from downstream, during the next 20 minutes after the beginning of hydropeaking.