The present contribution is motivated by O. Opitz's early investigations in cluster analysis. It presents clustering methods for quantitative data vectors, either by the classical k-means approach or by stochastic approximation. Its novelty resides in the fact that (a) we consider clustering criteria incorporating local regression models in various forms, (b) we introduec such models into the framework of Kohonen maps, thus obtaining regression-type self-organizing networks (SOMs) where neighbouring lattice points represent clusters with similar regression hyperplanes, and (c) we base our generalized SOM algorithms on discrete and continuous optimization criteria (e.g., the K-criterion). This will reveal the theoretical properties of the methods (performance, convergence etc.).