From January 1980 to December 1992, 655 cementless total knee arthroplasties with the PCA total knee system were done at the authors' institution. Thirty of these knees have been revised for aseptic loosening of either tibial or femoral components or both. Patients in the failed arthroplasty group were matched directly to a control group with well-functioning, stable and painless cementless total knee prostheses (Knee Society objective score, > 90 points). In this study, the authors analyzed any alignment differences between the 2 groups of knees. Special attention was given to the postoperative radiographs to evaluate for potential predictors of success versus failure. Multiple radiographic parameters were measured for each knee and included measures of frontal alignment, sagittal alignment, rotation, percent prosthetic coverage, and cortical thickness (denoted as the cortical-cancellous index). On univariate analysis, no individual radiographic parameter was significantly different in the failure group versus the control group. However, multivariate analysis indicated that certain combinations of abnormal parameters were significantly higher in the failed group than the control group. These parameters included measures of frontal and sagittal plane component alignment. The summated abnormal parameters per patient averaged 3.67 (range, 1-7) for the failure group and 1.80 (range, 0-7) for the control group.