OBJECTIVE To compare two plain radiographic methods for sensitivity to detect progression of patellofemoral osteoarthritis. METHODS Two sets of paired skyline and lateral knee radiographs from 54 hospital referred patients (108 knees) with knee osteoarthritis were taken an average of 31 months apart (range 12-40). Films were examined separately in random order by a single observer blind to patient identity and time order. Minimum joint space was measured by metered caliper; individual features of osteoarthritis were graded 0-3 using an atlas. RESULTS Intraobserver reproducibility assessed on 40 knees was to within +/- 0.5 mm for skyline lateral facet and +/- 0.7 mm for medial facet and lateral views. On the lateral view measured joint space decreased in 51% of knees but increased in 43%, with overall no significant mean group change with time (-0.2 mm, 95% confidence interval, 0.1 to -0.5). By contrast on the skyline view joint space decreased in at least one facet in 71% of knees, with significant decrease in mean joint space for both lateral facets (-0.4 mm, 95% CI, -0.2 to -0.6) and medial facets (-0.5 mm, 95% CI, -0.1 to -0.8). CONCLUSIONS It is possible to detect significant joint space loss with time on the skyline view that is not apparent on the lateral view. The skyline view should be the method of choice to detect progression of patellofemoral osteoarthritis.