Hand gesture identification is a complex problem, where more number of muscles will be involved even for a simple hand movement. Surface electromyography (sEMG) is an indicator of muscle activity and related to body movement and posture. In the recent past sEMG had been used with various statistical signal processing technique to identify different hand gestures, but since the hand actions require simultaneous muscle contractions reliability issues exist. Recently Blind source separation (BSS) techniques like Independent Component Analysis (ICA) had been used to tackle this problem. In this paper, a novel method is proposed to enhance the performance of ICA of sEMG by decomposing the signal into components originating from different muscles. First, we use FastICA algorithm to generate random mixing matrix, and the best mixing matrix is chosen based on the highest Signal to interference ratio(SIR) of mixing matrix. Pattern classification of the separated signal is performed in the second step with a back propagation neural network. The proposed model-based approach is able to overcome the ambiguity problems (order and magnitude problem) of BSS methods by selecting an apriori mixing matrix based on known hand muscle anatomy. Testing was conducted using several single shot experiments conducted with seven subjects. The results indicate that the system is able to classify six different hand gestures with 99% accuracy.