In the era of big data,bearing fault monitoring has the problem that it cannot realize the real-time processing of massive data processing and have the subjectivity about fault feature selection. In order to solve the above problems,a new bearing fault diagnosis method combining Compressed Sensing（ CS） and Multilayer Extreme Learning Machine（ ML-ELM） is proposed. This method first uses compressed sensing theory to obtain a small amount of data that can express characteristic information from a large amount of bearing monitoring data. Then multilayer extreme learning machine which improved by PSO use the measured value to obtain the classification information of bearing failure. This method greatly reduces the amount of data for bearing diagnostic signals and eliminate the step of feature selection in intelligent diagnosis. It makes full use of multilayer extreme learning machines to extract the characteristic information of bearing signals from a small number of observations.Experimental analysis show that this method can accurately identify faults at different positions and different damage levels,and provides a new method for bearing condition monitoring and fault diagnosis.