Using a method combining the projection pursuit method, the sensitivity and clustering to bearing status recognition of 24 characteristic indexes was studied. Using an inner ring fault as an example, a mathematical model was constructed to simulate fault signals, and 24 characteristic indexes were generated, then projected. The best projection direction matrix was put forward, and the projection distribution characteristic was studied. The range coefficient, coefficient of mean deviation, coefficient of dispersion, main axis of the relative coefficient and coefficient of mean value were proposed to study the sensitivity and clustering to different failure status of the 24 characteristic indexes projection. It can be validated based on of the inner ring fault test data publicly released on the Case Western Reserve University Bearing Data Center website. This method can be used to screen high quality characteristics for bearing fault diagnosis, which can guarantee prompt fault recognition and accurate diagnosis.