The aim of this study was to evaluate the interobserver reliability of parametric and non-parametric variables in the clinical assessment of hip and knee osteoarthritis (OA). Three rheumatologists examined 49 patients with different radiological stages of OA using different assessment tools such as a tape measure, a goniometer, a plurimeter and a hand-held pull gauge. The reliabilities of parametric variables calculated by analysis of variance (ANOVA) showed much higher values than the non-parametric ones calculated by Kendall's tau beta. The highest levels of correlation in hip OA between clinical functional tests and radiological changes were found for hip extension (r = 0.57; P < 0.01) and the Patrick sign (r = 0.54; P < 0.01) while in knee OA the highest correlations were found for knee circumference (r = 0.5; P < 0.01) and knee flexion (r = 0.035; P < 0.02). Knee muscle strength, as measured with a hand-held pull gauge, showed a high level of interobserver agreement (r = 0.79), but correlated poorly with radiological changes. In conclusion parametric variables of joint morphology as knee circumference of parametric variables of function as the Patrick sign should be preferred for assessing secondary endpoints in OA clinical trials.