Anterior cruciate ligament(ACL) injury of the knee is common and a serious ACL injury leads to ligament reconstruction surgery. Gait analysis is used to identify the result of surgery. The purpose of this study is to numerically evaluate and classify knee condition of patients through the chaos analysis. Experiments were carried out for 13 subjects (8 healthy subjects, 5 ACL deficient patients) walking on a treadmill. Sagittal kinematic data of the right lower extremity were collected by using a 3D motion analysis system. The recorded gait patterns were digitized and then coordinated by KWON3D. The largest Lyapunov exponent from the measured knee angular displacement time series was calculated to quantify local stability. It was found that the Lyapunov exponent becomes larger as the knee condition becomes worse. This study suggested a method of the severity of injury and the level of recovery. The proposed method discerns difference between healthy subjects and patients.