An empirical characterization of a family of condensing algorithms for the 1-NN rule with regard to the different Learning Vector Quantization schemes is presented. In particular, generalized prototype merging based on consistency on one hand and adaptive placement of a prespecified number of prototypes on the other, are considered. Both families of methods have advantages and drawbacks. Basically, LVQ methods tend to be more robust and efficient but they strongly depend on initialization and parameter setting while Consistency-based merging methods have no initialization and parameter setting but tend to be very dependent on the particular training data.