In this sensitivity analysis on a 3D model of a heterogeneous fluvial reservoir, two scenario orders have been considered. The first one focuses on the first-order heterogeneity (i.e. a fluvial belt with a 100% sand content), and the other one on the second-order heterogeneity accounting for the internal sedimentary fill within the fluvial belt (oxbow lakes). CO2 injections were simulated using THOUGH2, and the dynamic simulations show large variations of reservoir performances. The first-order heterogeneity generates a large spectrum of storage capacities ranging from 30 to 50 Mt, to be related to the natural connectivity variability between fluvial belts induced by the avulsion process. Considering second-order heterogeneity reduces the storage capacities by 30%, highlighting the importance of representing such objects in complex heterogeneous systems. Moreover, it increases the dissolution process, increasing by the way the storage efficiency. The CO2 plume extension and geometry is also estimated to be strongly dependent on the level of heterogeneity. Finally, trapping into poorly connected fluvial point bars affects strongly the storage capacity of the mobile CO2 as well as the pressure field.