Fuzzy classifiers are suited for pattern classification when there exist a large amount of imprecision, uncertainty and ambiguity in the patterns. One such fuzzy classifier is based on the possibilistic fuzzy membership function used for measuring the degree of class belongingness. However, the performance of possibilistic classifier depends heavily on the cluster parameters such as the 3-dB point and the parameter that controls the degree of fuzziness in the cluster. In this paper, we develop an iterative method for tuning these parameters so that the performance of the classifier is improved. The classifier considered in our work is a multi-prototype classifier and includes options for rejecting patterns that are ambiguous and/or do not belong to any class. In our proposed scheme, the slopes of the membership function are suitably varied via parameter tuning so that the membership of a pattern to a cluster in which it actually belongs is maximized while that to other classes are forced to be as small as possible. We evaluate our method using the Wisconsin Breast Cancer Dataset (WBCD). The results show that the recognition rate is improved by as much as 8% when the cluster parameters are tuned.