Bearing failure is one of the important reasons for machine accidents. To better identify fault types, this paper used a method which combines envelope sample entropy with decision tree gate limit value adaptive adjustment algorithm. We firstly decomposed the signal into several IMFs and selected the sample entropy of the envelope signal from the IMF which contains a number of diagnostic information, and then, the type of bearing failure was accurately determined by decision tree adaptive adjustment gate limit value.Analysis results showed that this method can not only reduce the computational complexity through the feedback, but also improve the recognition rate of bearing failure through the decision tree threshold adaptive adjustment. Comprehensive recognition rate can reach 96.75%.