The anterior cruciate ligament (ACL) plays an important role in controlling knee joint stability. The literature provides conflicting information on whether patients with ACL deficiency exhibits gait adaptations. The aim of this study is to investigate if the use of neural networks with a new pattern recognition-based method can differentiate gait patterns between ACL deficient (ACL-D) knee and contralateral ACL-intact (ACL-I) knee in patients with unilateral ACL deficiency. The proposed method is divided into two stages. In the training stage, kinematic and kinetic gait variables are measured and compared between the two lower extremities. Gait dynamics underlying gait patterns of ACL-D and ACL-I knees are locally accurately modeled and approximated by radial basis function (RBF) neural networks via deterministic learning theory. The derived knowledge of approximated gait dynamics is preserved in constant RBF networks. In the classification stage, a bank of dynamical estimators is constructed using the preserved constant RBF networks to represent the learned training gait patterns. By comparing the set of estimators with a test gait pattern, the generated average L1 norms of errors are taken as the difference and classification measure between the training and test gait patterns. Finally, experiments are carried out on forty-three patients to assess the effectiveness of the proposed method. By using the leave-one-out cross-validation style under normal and fast walking speed conditions, the correct classification rates for discriminating between ACL-D and ACL-I knees are reported to be 95.9% and 94.0% , respectively. Compared with other state-of-the-art methods, the results demonstrate that gait alterations in the presence of chronic ACL deficiency can be detected with superior performance and support the validity of the proposed method.