Changing the appearance of human gait due to the clothing variations makes the gait recognition problem much harder. In this paper, we study the gait signature by dividing the human body into very small fragments and investigate the effectiveness of these fragments when it combines one after another in step-wise. It has been demonstrated that dynamic parts shows more uncertainty thus for more informative in gait recognition. We have used more dynamic weighted entropy based gait representation in our parts definition as input gait features. The result of the proposed method outperforms others reported whole based and part based gait recognition approaches.