The purpose of this study was to demonstrate that specialized magnetic resonance imaging provides an accurate assessment of lesions of the articular cartilage of the knee. Arthroscopy was used as the comparative standard. Eighty-eight patients who had an average age of thirty-eight years were evaluated with magnetic resonance imaging and subsequent arthroscopy because of a suspected meniscal or ligamentous injury. The magnetic resonance imaging was performed with a specialized sequence in the sagittal, coronal, and axial planes. Seven articular surfaces (the patellar facets, the trochlea, the femoral condyles, and the tibial plateaus) were graded prospectively on the magnetic resonance images by two independent readers with use of the 5-point classification system of Outerbridge, which was also used at arthroscopy. Six hundred and sixteen articular surfaces were assessed, and 248 lesions were identified at arthroscopy. Eighty-two surfaces had chondral softening; seventy-five, mild ulceration; fifty-three, deep ulceration, fibrillation, or a flap without exposure of subchondral bone; and thirty-eight, full-thickness wear. To simplify the statistical analysis, grades 0 and 1 were regarded as disease-negative status and grades 2, 3, and 4 were regarded as disease-positive status. When the grades that had been assigned by reader 1 were used for the analysis, magnetic resonance imaging had a sensitivity of 87 per cent (144 of 166), a specificity of 94 per cent (424 of 450), an accuracy of 92 per cent (568 of 616), a positive predictive value of 85 per cent (144 of 170), and a negative predictive value of 95 per cent (424 of 446) for the detection of a chondral lesion. Interobserver variability was minimum, as indicated by a weighted kappa statistic of 0.93 (almost perfect agreement). With use of this readily available modified magnetic resonance imaging sequence, it is possible to assess all articular surfaces of the knee accurately and thereby identify lesions that are amenable to arthroscopic treatment.