In this paper, a fault diagnosis system based on discrete wavelet transform (DWT) and artificial neural networks (ANNs) is designed to diagnose different types of fault in gears. DWT is an advanced signal-processing technique for fault detection and identification. Five features of wavelet transform RMS, crest factor, kurtosis, standard deviation and skewness of discrete wavelet coefficients of normalized vibration signals has been selected. These features are considered as the feature vector for training purpose of the ANN. A wavelet selection criteria, Maximum Energy to Shannon Entropy ratio, is used to select an appropriate mother wavelet and discrete level, for feature extraction. To ameliorate the algorithm, various ANNs were exploited to optimize the algorithm so as to determine the best values for ‘‘number of neurons in hidden layer” resulted in a high-speed, meticulous three-layer ANN with a small-sized structure. The diagnosis success rate of this ANN was 100% for experimental data set. Some experimental set of data has been used to verify the effectiveness and accuracy of the proposed method. To develop this method in general fault diagnosis application, an example was investigated in cement industry. At first, a MLP network with well-formed and optimized structure (20:12:3) and remarkable accuracy was presented providing the capability to identify different faults of gears. Then this neural network with optimized structure is presented to diagnose different faults of gears. The performance of the neural networks in learning, classifying and general fault diagnosis were found encouraging and can be concluded that neural networks have high potential in condition monitoring of the gears with various faults.