Summary. We tested the reliability and construct validity of the Compatible magnetic resonance imaging (MRI) scale for the evaluation of haemophilic knees and ankles and compared the diagnostic performance of MRI and plain film radiographs. Sagittal and coronal gradient-echo 1.5-T MR images of knees (n = 22) and ankles (n = 23) were obtained from boys (age range 4–16 years; mean 11 years) in two centres (Toronto, n = 26; Europe, n = 19). The MR images were independently read by four blinded radiologists on two occasions. Number of previous joint bleedings and laboratory level of severity of haemophilia were the reference standards for imaging assessment. Both components of the MRI scale demonstrated high inter- and intrareader intraclass correlation coefficients (progressive (P) scale, 0.91 and 0.94; additive (A) scale, 0.81 and 0.92 respectively). The correlation between the osteochondral domain of the MRI scale and patient's age was moderate. Otherwise, correlations between A- and P-scales and clinical laboratory measurements were weak. The areas under the curve (AUCs) used for discrimination of disease severity were similar for the A- and P-scales (AUCs used for mild disease, A-scale, 0.72 ± 0.07; P-scale, 0.69 ± 0.08; P = 0.23; AUCs for severe disease, A-scale, 0.93 ± 0.05; P-scale, 0.87 ± 0.08; P = 0.05). No differences were noted between the AUCs of the MRI and radiographic scales used for discrimination of late osteoarticular changes; MRI scales performed better for discrimination of early changes. In conclusion, both MRI scales demonstrated excellent reliability, poor convergent validity, and moderate and excellent validity for discrimination of mild and severe diseases respectively. Compared with radiographic scores, the MRI scales performed better for discrimination of early osteoarticular changes.