Abstract: Breast cancer is one of the popular cancers in women and is considered one of the popular causes of death. Earlier detection and diagnosis may save lives and make efficient of life. In this study, a new method for breast cancer diagnosis is proposed. The proposed method consists of three stages: the first divides dataset to two clusters using kernel k-means clustering, the second minimizes features by applying feature selection algorithm on each cluster and the third collects resulting feature from each cluster together and measures the quality using different classifiers. The proposed approach is evaluated using datasets for breast cancer: Breast cancer wisconsin diagnostic dataset "WDBC" get from UCI machine learning repository. The performance of the proposed method is evaluated by measuring accuracy, sensitivity, specificity, mean squared error and time. The experiments are done with three classifiers Naive Bayes "NB", Multilayer Perceptron "MLP" and decision tree J48.