Downstream of hydroelectric storage dams, the rivers suffer from rapid, frequent and intense flow variations linked to the power production constraints. Subjected to this hydropeaking phenomenon, the fish population suffers from a known decline. In the research of mitigation solutions, shelters in riverbanks are a protection measure against excessive flow velocities. This project gives the design basis for such shelters. The research process is based on experimentations with fish swimming in a canal built on purpose with a lateral embayment. The principle is to generate hydrodynamic conditions that are uncomfortable in the canal on one hand and attracts the fish to the shelter on the other hand. The significant parameter is the flow velocity distribution in the intersection between the main channel and the lateral fish shelter. The frequentation rate of fish in the shelter can be considered as the performance indicator. With a length of 12 m and a width of 1.2 m, the canal is supplied with freshwater from a river. At mid-length, the canal holds a one side enlargement of 1.2 m, shaped like a rectangular gap of 2 m long. For each test, after a steady flow of 10 l/s with a mean velocity of 0.1 m/s, 10 to 20 juvenile brown trout are suddenly exposed during 3 hours to a hydropeaking flow of 220 l/s. At that time, the mean velocity in the canal is of 1 m/s. The fish position is observed every 20 minutes and their movements are recorded continuously in the shelter sector. The first tests have been conducted with a basic rectangular shelter configuration. The frequentation rate was only of 35%. This rate has doubled by introducing a deviation groyne that enforced a water exchange between the canal and the shelter. Following this experiment, the position and the orientation angle of the groyne have been systematically varied with the purpose of maximising the frequentation rate. This way, 12 configurations have been tested 3 times each. The flows transiting in the shelter are calculated using a 2D numerical simulation. By analysing the fish positioning, the mean frequentation rate is determined for each configuration. The maximum rate is of 87%, the mean is of 75% and the standard deviation is of 8%. Using video pictures treatment, the preferential fish trajectories are updated and the fish entrance distribution is determined for each configuration at the interface between the shelter and the channel. On the same interface, ultrasonic velocity profiles (UVP) have been measured for the 12 configurations. By associating the fish entries distribution with the velocities distribution, the preferences are globally established. The entrances happen mostly from downstream of the derivation groyne, in a flow shear zone where the velocities are in a range of 0.0 to 0.2 m/s. Finally, the analysis of all the results confirmed the representativity of the configuration that gives the best frequentation rate, for the development of the experimental shelter's prototype to be realised in rivers. It is composed of a groyne shaped like an equilateral triangle. Finally, recommendations are given for the execution of an experimental prototype shelter to be realized in a river for validation tests.