Experiments in an asymmetric compound channel with a strongly converging floodplain are presented. The results are compared to predictions given by two different 1D programmes. Each programme uses a particular method to account for the momentum transfer that occurs in compound channels, due to the velocity gradient between the main channel and the floodplain. The first one is based on the DEBORD formula, which is an empirical correction of the Divided Channel Method. The other one is based on the Exchange Discharge Model, which computes momentum and discharge exchanges at the interface between the two subsections. In the studied situation, where significant three-dimensional phenomena are detected, both numerical models exhibit some weaknesses. Particularly, it is difficult to get simultaneously correct results for the water depth, the discharge distribution in the compound section, and the subsection mean velocities. The analysis of the momentum balance of the flow provides information on the additional phenomena that exist in the experimental situation. These phenomena should be taken into account in the 1D models when such strong conveyance decrease is considered.