This paper aims to demonstrate the traditional maximum entropy fuzzy C-means clustering algorithm (MEFCM) applies to spherical or oval-shaped clusters only. In order to solve the confusion and highly relevant data distribution division of this difficult situation, it introduces Mercer kernel function, so that the original features which do not show can stand out and make the clustering effect better. However, in practical, the majority of sample sets are exist importance (weighting) of different phenomena. The main focus are the samples of different importance to design of each data weighting and in order to overcome the sensitivity of weakness of the initial cluster centers. This paper presents an optimization of the initial cluster centers weighted kernel maximum entropy fuzzy clustering algorithm (WKMEFCM). Experiments show that compared with the MEFCM, the clustering result is more stable, accurate and the clusters division effect is better.