This study tested whether the peak external knee adduction moments during walking in subjects with knee osteoarthritis (OA) were correlated with the mechanical axis of the leg, radiographic measures of OA severity, toe out angle or clinical assessments of pain, stiffness or function. Gait analysis was performed on 62 subjects with knee OA and 49 asymptomatic control subjects (normal subjects). The subjects with OA walked with a greater than normal peak adduction moment during early stance (p = 0.027). In the OA group, the mechanical axis was the best single predictor of the peak adduction moment during both early and late stance (R = 0.74, p < 0.001). The radiographic measures of OA severity in the medial compartment were also predictive of both peak adduction moments (R = 0.43 to 0.48, p < 0.001) along with the sum of the WOMAC subscales (R = −0.33 to −0.31, p < 0.017). The toe out angle was predictive of the peak adduction moment only during late stance (R = −0.45, p < 0.001). Once mechanical axis was accounted for, other factors only increased the ability to predict the peak knee adduction moments by 10–18%. While the mechanical axis was indicative of the peak adduction moments, it only accounted for about 50% of its variation, emphasizing the need for a dynamic evaluation of the knee joint loading environment. Understanding which clinical measures of OA are most closely associated with the dynamic knee joint loads may ultimately result in a better understanding of the disease process and the development of therapeutic interventions.