When comparing Self-Organizing Feature Map (SOFM) learning with the generalized Lloyd algorithm for vector quantizer design, it becomes apparent that an implicit goal of SOFM learning is to reduce the summed distortion between the input vector and all vectors in the neighborhood of the closest SOFM weight vector. We present appropriate neighborhood-based distortion measures for SOFM and derive a variant of the generalized Lloyd algorithm which, unlike traditional SOFM learning, monotonically reduces neighborhood distortion. We show that topology preservation naturally results from minimization of neighborhood distortion, and such distortion can be used to quantify the degree to which topology is preserved in SOFMs. We also report on the use of the topology preserving properties of SOFMs for speaker independent isolated digit recognition. SOFMs are found to be effective for input pattern normalization, pattern recognition, and feature integration.