In this paper a connectionist clustering strategy is presented for segmenting color images. First the local peaks in the 3-D R,G,B histogram are located. Then using these as the prototypes other patterns are classified to one of them. The prototype selection and classification networks have been analyzed. The prototype selection method employs only neuronal dynamics and therefore is faster than existing clustering neural networks. The classification network takes into account the distribution of the data and hence is less prone to misclassifications. Experimental results obtained by applying the network for segmenting one color image is presented. © 1992 IEEE.