Magnetic resonance imaging (MRI) scans were performed on 55 bladder cancer patients on whom clinicopathological staging was available from transurethral resection and cystectomy specimens. The overall accuracy of MRI scanning in this group was 84%, although true concordance rates are debatable without open surgical correlation. In the subgroup of 25 patients who had accurate open surgical correlation (from cystectomy, laparotomy or post mortem) the concordance rate was 76% with MRI. Errors occurred mainly in the T3 group of tumours, with 2 being overstaged and 2 being understaged out of a total of 12 in the open surgical correlation group (66% accuracy). Difficulties were also encountered in staging tumours at the bladder base, with an error rate of 22% (2 of 9) for T4 tumours in this area. With regard to lymph node staging there was a 100% (5 of 5) specificity in defining pathologically involved nodes but there was a false negative rate of 15% (3 of 19). Although it has many advantages over CT scanning, MRI produces a significant error rate in terms of over- and under-staging invasive tumours. There are difficulties associated with detecting minimal involvement of adjacent organs and lymph nodes as well as determining the exact depth of muscle penetration. Improvements may come in the future with the use of contrast enhancement agents such as gadolinium as well as more advanced machines.