Once the marketing analyst wishes to determine Competitive Market Structure (CMS) he or she may choose among a set of data reduction techniques (such as cluster analysis of multidimensional scaling) that produce discrete (non-spatial) or spatial configurations of competing products/brands with respect to their observed substitutionality as seen by consumers. Since input data are organized in a two- or multi-mode manner (e.g. consumers' brand choice probabilities or preference data, attribute ratings, etc.), CMS always has to deal with consumer heterogeneity. Therefore, advanced methods like the MULTICLUS procedure by DeSarbo, Howard and Jedidi (1991) combine the two interrelated tasks of CMS and segmentation analysis into one single model. Performance of the Self-Organizing (Feature) Map (SOM) methodology as originally proposed by Kohonen (1982) is compared to MULTICLUS solutions in such a combined CMS/segmentation context. As a special variant of artificial neural network models SOMs also perform simultaneous clustering and topological representation of data vectors and thus seem to be well-prepared for multi-mode data analysis. As input data household-level choice probabilities for (potential) rival brands as derived from panel data are used. Unlike the maximum likelihood based MULTICLUS procedure, the SOM methodology arrives at a non-linear topological projection of high-dimensional input data onto a two-dimensional discrete map through an adaptive training process. Once SOM training is completed, the latter can be identified via households' assignments (a posteriori segmentation) to fuzzy partitions representing distinctive patterns of brand competition (segment-specific CMS). Relative similarities between MULTICLUS and SOM solutions are investigated and performance requirements are discussed.