This paper is devoted to the development of feature extraction methods for speaker recognition. A new approach based on histograms of mel-frequency cepstral coefficient (MFCC) distributions to calculate feature vectors for voice samples is proposed. The resulting vectors appear to be independent of original voice sample length and have relatively small sizes. They incorporate the spread of unique vocal tract related characteristics which can be used as distinctive features for recognition. This approach of voice recognition is implemented in a software module developed for MATLAB environment. A support vector machine method and Voicebox speech processing toolbox for MATLAB are utilized. Results of the developed module test runs are obtained and reported. A comparison of test results with results of traditionally used feature vector based techniques of speaker recognition shows relatively low rates of false acceptance and false match for the proposed approach. Feature vectors based on MFCC distributions can be effectively used in real world voice recognition systems.