Finger recovery is much harder than other parts on the upper limbs, because finger recovery movement has several key problems need to overcome, including high precision of movement, high control resolution requirements, variable data with different person, as well as the fuzzy signal during the movement. In order to overcome the difficulties, a new scheme of finger recovery is presented in the paper based on symmetric rehabilitation. In the paralyzed hand side, a mechanical exoskeleton hand is designed and simulated to provide skeletal traction, while in the regular hand side, the curve magnitude of every joint during movement is detected. Then the hand motion is analyzed and recognized using Multi-class SVM. Many candidates were chosen to perform the experiment, and the data produced by the candidates were divided the training parts and recognition parts. Experiments shows that the Multi-class SVM is effective and practical for classification and recognition, and could be helpful in the finger recovery process.