The use of biometric information has been known widely for both person identification and security application. Each person can be identified by the unique characteristics of one or more of person biometrics. One of the biometric characteristics of that a person can be identified by his voice. In this research, we are interested in studying the effect of proper features that are extracted from discrete wavelet transformation. The research illustrates the effect of using different percent of each level (instead of all features within each level) on speaker identification accuracy. Feed-forward Back propagation (BP) neural network and learning vector quantization (LVQ) are used as classifiers. The neural networks trained with feature extracted from different levels of discrete wavelet transform (one level at a time). It was found that level 3, 5 are the best levels to extract features. 50% of each level is enough for the recognition process, at which the recognition rate reaches 96% with BP NN neural network. The system was trained and tested using cross-validation.