OBJECTIVES: No previous study has adequately demonstrated the test-retest reliability of the Short-Form McGill Pain Questionnaire, yet it is increasingly being used as a measure of pain. This study evaluates the test-retest reliability in patients with osteoarthritis. METHODS: A prospective, observational cohort study was undertaken using serial evaluation of 57 patients at 2 time points. A sample of patients awaiting primary hip or knee joint replacement surgery were recruited in clinic or via mail (mean age 64.8 years). Short-Form McGill Pain Questionnaires were delivered by mail 5 days apart, and a supplementary questionnaire was completed on the second occasion to explore if the patients' pain report had remained stable. RESULTS: The intraclass correlation coefficient was used as an estimate of reliability. For the total, sensory, affective, and average pain scores, high intra-class correlations were demonstrated (0.96, 0.95, 0.88, and 0.89, respectively). The current pain component demonstrated a lower intraclass correlation of 0.75. The coefficient of repeatability was calculated as an estimation of the minimum metrically detectable change. The coefficients of repeatability for the total, sensory, affective, average, and current pain components were 5.2, 4.5, 2.8, 1.4 cm, and 1.4, respectively. DISCUSSION: Problems of adequate completion of the Short-Form McGill Pain Questionnaire were highlighted in this sample, and supervision via telephone contact was required. Patients recruited in clinic who had practiced completing the Short-Form McGill Pain Questionnaire demonstrated fewer errors than those recruited by mail. The Short-Form McGill Pain Questionnaire was demonstrated to be a highly reliable measure of pain. These results should not be generalized to a more elderly population, as increasing age was correlated with greater variability of the sensory component scores.