Geological heterogeneities directly control underground flow. In channelized sedimentary environments, their determination is often underconstrained: it may be possible to observe the most recent channel path and the abandoned meanders on seismic or satellite images, but smaller-scale structures are generally below image resolution. In this paper, reconstruction of channelized systems is proposed with a stochastic inverse simulation reproducing the reverse migration of the system. Maps of the recent trajectories of the Mississippi river were studied to define appropriate relationships between simulation parameters. Measurements of curvature and migration vectors showed (i) no significant correlation between curvature and migration offset and (ii) correlation trends of downstream and lateral migration offsets versus the curvature at half-meander scale. The proposed reverse migration method uses these trends to build possible paleo-trajectories of the river starting from the last stage of the sequence observed from present-day (satellite or seismic) data. As abandoned meanders provide clues about the paleo-locations of the river, they are integrated time step by time step during the reverse simulation process. We applied the method to a satellite image of a fluvial system. Each of the different resulting geometries of the system honored most of the available observations and presented meandering patterns similar to the observed ones.