We describe a new approach to the problem of reconstructing 3-D branching objects from a small number of projections. Given the object-to-image transformations and an assumption of structural connectivity, we describe how to elucidate the imaged structure from the multitude of artifacts resulting from all possible triangulations. The method involves generating two or more intermediate reconstructions from different pairs of projections, then comparing structural similarities/differences to identify the artifacts. Since the intermediate reconstructions are assured to be connected (i.e. not fragmented), it becomes possible to refine portions of the structure that previously were considered ambiguous. Simulations were performed using a mathematically defined test object having 10 branches. In every case, the true structure was able to be distinguished from the artifacts using as few as three projections.