Traditional feature selection methods don’t consider the pattern feature of data, which leads to the performance degradation. ReliefF is a more effective method for feature selection, but the weight of feature can fluctuate with samples and the method cannot remove redundant feature. Therefore, the concept of authentic distance and a new method used to feature selection are proposed based on the pattern feature of data, and the feature selection and experiments of sintering working conditions recognition are operated with the data of alumina rotary kiln sintering process as experimental data. The experimental results show that the proposed method can effectively remove unrelated features and greatly improve recognition rate.