The value of plain radiographs, digital subtraction arthrography and radionuclide arthrography was analysed in 23 cases of failed total knee arthroplasty. The preoperative diagnosis was compared with the intraoperative assessment. Sensitivity, specificity and the positive and negative predictive value for assessing a loose component were determined separately for the femoral and tibial components. At revision we found 13 loose femoral and 12 loose tibial implants. In eight cases both components were unstable. Plain radiography had a sensitivity of 77% for loosening of the femoral and 83% for the tibial component; digital subtraction arthrography 77% for the femoral and 8% for the tibial component and radionuclide arthrography 31% and 8%. The specificity for plain radiography was 90% for the femoral and 72% for the tibial implant. For subtraction arthrography it was 50% and 82% and for subtraction arthrography 70% and 82%. Radiography had the highest positive and negative predictive values for both components compared with the other two techniques. As a diagnostic tool to detect implant loosening, plain radiography is the most effective in this study. Subtraction arthrography and radionuclide arthrography are not suitable for use as routine methods for detection of total knee arthroplasty loosening.