Successful localization of sound sources in reverberantenclosures is an important prerequisite for many spatialsignal processing algorithms. We investigate the use of a weightedfuzzy ??-means cluster algorithm for robust source localizationusing location cues extracted from a microphone array. In orderto increase the algorithm’s robustness against sound reflections,we incorporate observation weights to emphasize reliable cuesover unreliable ones. The weights are computed from local featurestatistics around sound onsets because it is known that theseregions are least affected by reverberation. Experimental resultsillustrate the superiority of the method when compared withstandard fuzzy clustering. The proposed algorithm successfullylocated two speech sources for a range of angular separations inroom environments with reverberation times of up to 600 ms.