Advance in machine learning techniques for machine condition monitoring is now an active field of interest in modern industry for predicting the future health of machines. In this paper, Artificial Neural Networks (ANNs) technique is investigated by using the existing datasets of ball bearing fault to classify the fault location and severity. To enhance the accuracy, the utilization of statistical features and spectral features are used in pre-processing with the proposed method of cascade ANNs. Motor vibration data is collected by using accelerometers attached to the housing in the drive end and fan end. To diagnose the faults in the rolling bearing, the vibration data is pre-processing by using Fast Fourier Transform (FFT) and other statistical features such as standard deviation, skewness and kurtosis etc.  Then the feature data is fed into cascade ANN_1 and ANN_2 to classify the fault location and severity. The objective of this research is to solve the problem of poor features such as slippage frequency component in FFT response and inappropriate statistical feature datasets to help increase in accuracy and efficiency. The results show that the proposed method gives more accuracy when the poor features are existed and guarantee the accuracy near to 100%.