In this paper, an improvement of experimental for pattern recognition of SEMG signals of hand gesture using spectral estimation and neural network is proposed. Proposed spectral estimations are the Covariance method and the Linear Vector Quantization (LVQ) is chosen for Neural Network classification. The raw SEMG signals been captured from SEMG Amplifier and the Auto Regressive (AR) Covariance returned the power spectral density (PSD) magnitude squared frequency response. The 4 channels of AR data will be combined and a fine tuning step by using LVQ will then incorporate for pattern classification. The detail of the experiment and simulation conducted described here to verify the differentiation and effectiveness of combined channels PSD method for SEMG pattern classification of hand gesture.