OBJECTIVE To (a) develop further logically derived line drawing atlases (LDAs) for grading radiographic knee osteoarthritis (OA); and (b) determine which is superior using metrological criteria. METHODS A series of LDAs (-3 to +3, -4 to +4, and -5 to +5) were produced by (a) incorporating additional grades for osteophyte and joint space width (JSW) above the 0-3 pilot LDA, over an equivalent range of disease; and (b) adding negative grades for JSW. 121 sets of bilateral knee radiographs (standing, anteroposterior plus flexed skyline), plus serial views of 68 tibiofemoral joints (TFJs) and 36 patellofemoral joints were scored twice by one observer for each LDA. Minimum JSW of 50 radiograph sets was directly measured and awarded a categorical grade dependent upon the boundaries of each LDA grade. Time taken to grade 30 randomly selected knee radiograph sets was measured. RESULTS Intraobserver reproducibility was similar for all LDAs, (weighted kappa: JSW = 0.85-0.87; osteophyte = 0.77-0.79), with no deterioration with increasing grades. Criterion validity favoured the -5 to +5 LDA, which was also quickest to use. All atlases showed similar responsiveness (standardised response mean: medial TFJ JSW = 0.78-0.83; medial femoral osteophyte = 0.61-0.73), with most sites compromised by small sample size, little change in score, and high variation between subjects. CONCLUSIONS A set of LDAs was created illustrating the full range of normality/abnormality likely to be encountered in a community study of knee pain or OA. Despite superior validity and equivalent reproducibility, improved responsiveness of the -5 to +5 LDA was not confirmed.