[en] The need to explore structure in high-dimensional clouds of data points that may concentrate near (possibly nonlinear) manifolds of lower dimension led to the current development of three new approaches. The first is a computer-graphic system (PRIM'79) that facilitates interactive viewing and manipulation of an ensemble of points. The other two are automatic procedures for separating a cloud into more manageable pieces. One of these (BIDEC) performs successive partitioning of the cloud by use of hyperplanes; the other (Cake Maker) explores expanding sequences of neighborhoods. Both procedures provide facilities for examining the resulting pieces and the relationships among them