Abstract Breast cancer is the most commonly witnessed cancer amongst women around the world. Computer aided diagnosis (CAD) have been playing a significant role in early detection of breast tumors hence to curb the overall mortality rate. This work presents an enhanced empirical study of impact of dominance-based filtering approach on performances of various state-of-the-art classifiers. The feature dominance level is proportional to the difference in means of benign and malignant tumors. The experiments were done on original Wisconsin Breast Cancer Dataset (WBCD) with total nine features. It is found that the classifiers' performances for top 4 and top 5 dominant-based features are almost equivalent to performances for all nine features. Artificial neural network (ANN) is come forth as the best performing classifier among all with accuracies of 98.9% and 99.6% for top 4 and top 5 dominant features respectively. The error rate of ANN between all nine and top 4 &5 dominant features is less than 2% for four performance evaluation parameters namely sensitivity, specificity, accuracy and AUC. Thus, it can be stated that the dominance-based filtering approach is appropriate for selecting a sound set of features from the feature pool, consequently, helps to reduce computation time with no deterioration in classifier's performance.