Previous studies have demonstrated the importance of joint angle errors mainly due to skin artifact and measurement errors during gait analysis. Joint angle errors lead to unreliable kinematics and kinetic analyses in the investigation of human motion. The purpose of this paper is to present the Joint Averaging Coordinate System (JACS) method for human gait analysis. The JACS method is based on the concept of statistical data reduction of anatomically referenced marker data. Since markers are not attached to rigid bodies, different marker combinations lead to slightly different predictions of joint angles. These different combinations can be averaged in order to provide a "best" estimate of joint angle. Results of a gait analysis are presented using clinically meaningful terminology to provide better communication with clinical personal. In order to verify the developed JACS method, a simple three-dimensional knee joint contact model was developed, employing an absolute coordinate system without using any kinematics constraint in which thigh and shank segments can be derived independently. In the experimental data recovery, the separation and penetration distance of the knee joint is supposed to be zero during one gait cycle if there are no errors in the experimental data. Using the JACS method, the separation and penetration error was reduced compared to well-developed existing methods such as ACRS and Spoor & Veldpaus method. The separation and penetration distance ranged up to 15 mm and 12 mm using the Spoor & Veldpaus and ACRS method, respectively, compared to 9 mm using JACS method. Statistical methods like the JACS can be applied in conjunction with existing techniques that reduce systematic errors in marker location, leading to an improved assessment of human gait.