Multipath clustering is an essential concept for the most well-known geometry based (stochastic) channel models as from 3GPP, COST and WINNER. They assume that multipath components arrive/depart in clusters with similar properties within the considered parameter domains. Therefore, reliable and accurate clustering results are essential. Within this contribution a new approach to initialize a K-Means clustering algorithm for widely automated data processing is proposed. The influence of different processing parameters on the clustering result is discussed and multipath cluster statistics from two urban macro cell environments are summarizing this contribution.