Purpose: The aim of this prospective study was to compare routine MRI scans of the knee at 1.5 and 3 T obtained in the same individuals in terms of their performance in the diagnosis of cartilage lesions. Methods: One hundred patients underwent MRI of the knee at 1.5 and 3 T and subsequent knee arthroscopy. All MR examinations consisted of multiplanar 2D turbo spin-echo sequences. Three radiologists independently graded all articular surfaces of the knee joint seen at MRI. With arthroscopy as the reference standard, the sensitivity, specificity, and accuracy of 1.5- and 3-T MRI for detecting cartilage lesions and the proportion of correctly graded cartilage lesions within the knee joint were determined and compared using resampling statistics. Results: For all readers and surfaces combined, the respective sensitivity, specificity, and accuracy for detecting all grades of cartilage lesions in the knee joint using MRI were 60, 96, and 87 % at 1.5 T and 69, 96, and 90 % at 3 T. There was a statistically significant improvement in sensitivity (p < 0.05), but not specificity or accuracy (n.s.) for the detection of cartilage lesions at 3 T. There was also a statistically significant (p < 0.05) improvement in the proportion of correctly graded cartilage lesions at 3 T as compared to 1.5 T. Conclusion: A 3-T MR protocol significantly improves diagnostic performance for the purpose of detecting cartilage lesions within the knee joint, when compared with a similar protocol performed at 1.5 T. Level of evidence: III.