A prognostic approach to defining chronic pain has been proposed as an alternative to traditional definitions based on retrospective duration of pain. While this new approach performs well in low back pain (LBP), headache and orofacial pain, it is not known whether it translates to regional pain syndromes with an underlying pathological component, such as osteoarthritis (OA). We investigated the performance of this approach in a population-based cohort of older adults reporting knee pain, with a spectrum of radiographic knee OA. 676 adults (50 years+) attended a research clinic and were followed up at 18 months and 3 years. Risk scores were calculated using pain intensity, pain duration, pain-related activity, number of pain sites and depressive symptoms, measured at baseline and at 18 months. These scores were used to determine the probability of future clinically significant knee pain, defined as Chronic Pain Grade II-IV, at 18 months and at 3 years using logistic regression. Cut-points on the risk score were applied to determine groups at intermediate (probability >or=0.2), possible (>or=0.5) and probable (>or=0.8) risk of clinically significant knee pain. Discriminative ability of the risk scores, determined by area under the ROC curve, was high (0.78-0.82), varied little by radiographic severity and was superior to pain duration alone. The derived cut-points suggested a lower threshold for each of the risk groups than the previous LBP work. This prognostic approach to defining chronic pain appears to translate well to knee pain. Different cut-points for defining risk groups may be needed for different pain syndromes.