In machine learning or data mining research area, clustering is definitely an active topic and has drawn a lot of attention for its significance in practical applications, such as image segmentation, data analysis, text mining and so on. There have been a great number of clustering algorithms derived from different points of view. K-means is widely known as a straightforward and fairly efficient method for solving unsupervised learning problems. Due to its inherent weaknesses in some cases, many enhancements have been made for it. Soft k-means algorithm is one of them. In this article, we propose an entropy based soft k-means clustering method which utilizes the entropy and relative entropy information from data samples to guide the training process, for reaching a better clustering result.