A new feature subset selection algorithm using class association rules mining is proposed in this paper. Firstly, the algorithm mines rules with features as antecedences and class attributes as consequences. Then, it selects the strongest rules one by one, and all the rules’ antecedences make up of the selected feature subset. Experimental results on 10 real data sets show that the proposed algorithm produces a remarkable advantage in enormously reducing the number of the features while keeping quite high classification accuracy compared with other existing seven algorithms. This algorithm can offer an available preprocess technique for pattern recognition, machine learning and data mining.