BACKGROUND: Magnetic resonance (MR) imaging is an accurate imaging modality of the knee. The role of MR in clinical practice has not been precisely defined, largely due to the relative expense of the technique. METHODS: For each new patient with a knee problem who was referred for MR, a provisional diagnosis was made together with a level of certainty regarding the diagnosis. The waiting time for the scan was recorded. To assess clinical usefulness the MR diagnosis was compared with the provisional diagnosis and classified according to the following descending order of value: unexpected negative (no intra-articular pathology), confirmatory negative, unexpected positive or confirmatory positive. To assess accuracy of the MR diagnosis, the operative diagnosis was compared to the MR diagnosis in those patients who underwent arthroscopy. RESULTS: Fifty-two per cent of scans were assessed as being very useful and a further 20% were assessed as being moderately useful. Magnetic resonance had a 95% accuracy for medial meniscal tears, 91% accuracy for lateral meniscal tears, and 98% accuracy for anterior cruciate ligament tears, similar to previously reported studies. The diagnostic arthroscopy rate in the patients who underwent MR scanning was similar to that in patients for whom the surgeon was more confident about the diagnosis and who therefore did not undergo MR scanning. The diagnostic arthroscopy rate could have been reduced if surgery had not been performed in 14 patients who had a negative MR scan. CONCLUSIONS: There is a role for selective use of MR in the assessment of knee conditions. In particular, MR can be used to reduce the diagnostic arthroscopy rate.