Joint space width (JSW) and narrowing (JSN) measurements on radiographs are currently the best way to assess disease severity or progression in hip osteoarthritis, yet we lack data regarding the most accurate and sensitive measurement technique. This study was conducted to determine the optimal radiograph and number of readers for measuring JSW and JSN. Fifty pairs of radiographs taken three years apart were obtained from patients included in a structure modification trial in hip osteoarthritis. Three radiographs were taken with the patient standing: pelvis, target hip anteroposterior (AP) and oblique views. Two trained readers, blinded to each other's findings, time sequence and treatment, each read the six radiographs gathered for each patient twice (time interval > or = 15 days), using a 0.1 mm graduated magnifying glass. Radiographs were randomly coded for each reading. The interobserver and intraobserver cross-sectional (M0 and M36) and longitudinal (M0-M36) reproducibilities were assessed using the intraclass coefficient (ICC) and Bland-Altman method for readers 1 and 2 and their mean. Sensitivity to change was estimated using the standardized response mean (SRM = change/standard deviation of change) for M0-M36 changes. For interobserver reliability on M0-M36 changes, the ICCs (95% confidence interval [CI]) were 0.79 (0.65-0.88) for pelvic view, 0.87 (0.78-0.93) for hip AP view and 0.86 (0.76-0.92) for oblique view. Intraobserver reliability ICCs were 0.81 (0.69-0.89) for observer 1 and 0.97 (0.95-0.98) for observer 2 for the pelvic view; 0.87 (0.78-0.92) and 0.97 (0.96-0.99) for the hip AP view; and 0.73 (0.57-0.84) and 0.93 (0.88-0.96) for the oblique view. SRMs were 0.61 (observer 1) and 0.82 (observer 2) for pelvic view; 0.64 and 0.75 for hip AP view; and 0.77 and 0.70 for oblique view. All three views yielded accurate JSW and JSN. According to the best reader, the pelvic view performed slightly better. Both readers exhibited high precision, with SRMs of 0.6 or greater for assessing JSN over three years. Selecting a single reader was the most accurate method, with 0.3 mm precision. Using this cutoff, 50% of patients were classified as 'progressors'.