Prediction of remaining service life of the ball bearings, based on self-organizing map (Self organizing map,SOM) and BP (Back propagation,BP) two kinds of neural networks, proposed a new method for predicting remaining life of ball bearing system. Comparative analysis of advantages and disadvantages of different bearing recession indicators, using three sets of indicators of recession and three sets of time domain frequency domain recession indicators, including a new set of indicators, train the self organizing map neural network. SOM-will be derived from the smallest quantization error (Minimum quantization error,MQE) as an indicator of a new recession, set bearing the performance database. Ball bearing the recession, training of BP neural network according to the expiration time of the value of the right technology, the successful development of a prediction model of remaining life. Results showed that, the programme is far superior to the industry commonly used L10 life.