This paper presents a system for real-time classification of hand gestures through surface eclectromyographic signals. Support vector machine was applied to identify hand gesture intention based on extracted features of six muscles and discriminate up to 10 different hand gestures. Absolute Teager-Kaiser Energy was applied to detect the onsets and terminal points of the hand gestures. The results showed that the classifier with this algorithm has a good performance in discriminating most gestures. The real-time classification performance differed between subjects. In addition, removing peak values out of the raw signals could further improve the real-time classifying reliability, though the time consumption increased, leading to an increased time delay for gesture classification. The proposed classifiers showed better performance in real-time gesture classification, and would improve the reliability of controlling wearable robotic hand.