In traditional FCM clustering algorithm each feature is supposed to have equal importance. Considering different feature with different importance, this paper presented an improved FCM algorithm with adaptive weight for features of each cluster, named AWFCM. In the iterative AWFCM process, to identify the importance of features of each cluster, the weight for feature is computed dynamically based on the variance of the within cluster distances of the feature, and the new weights are used to calculate the cluster memberships of objects in next iteration effectively. Moreover, for the reason that in traditional FCM the features with wider variation range have greater impact on the clustering result even if they are less important, AWFCM introduce an method to normalize the clustering data between 0 and 1 in order to eliminate the over effect of the features with wider variation range. And then, based on four real data sets from UCI, the experiments demonstrated the AWFCM algorithm outperformed the FCM algorithm.