This paper describes the features of KIN-nav, a navigation system specifically dedicated to intra-operative evaluation of knee laxity, and assesses the reliability of the system during surgery. The acquisition protocol for its intra-operative use, the original user interface, and the computational methods for elaboration of kinematic data are described in detail. Moreover, an extensive and specific validation of the system was performed in order to evaluate its intra-operative performance and usability. KIN-nav's reliability and accuracy were analyzed in a series of 79 patients undergoing ACL reconstruction. The intra-surgeon repeatability computed for ACL-deficient and reconstructed knees at different flexion angles was less than 0.6 degrees for varus-valgus (VV) rotation, less than 1 mm for AP translation, and less than 1.6 degrees for IE rotation. The inter-surgeon repeatability is less than 2 degrees for VV rotation, 5 degrees for internal-external rotation, and less than 3 mm for AP translation. The proposed method was fast (requiring an additional 10 minutes of surgical time on average), required only a short learning period (5 cases), was minimally invasive, and was robust from the numerical perspective. Our system clearly shows that the use of navigation systems for kinematic evaluation provides useful and complete information on the knee state and test performance, and is simple and reliable to use. The good repeatability in manual kinematic tests is an improvement on the present ability to discriminate knee kinematics intra-operatively, and thus offers the possibility of better discrimination between knee pathologies and the prospect of new surgical applications.