We assessed the efficacy of intraoperative frozen-section histology in detecting infection in failed arthroplasties in 106 hips and knees. We found inflammatory changes consistent with infection (an average of one or more neutrophil polymorphs or plasma cells per high-power field in several samples) in 18 cases; there was a significant growth on bacterial culture in 20 cases. Compared with the bacterial cultures, the frozen sections provided two false-negative results and three false-positive results (sensitivity, 90%; specificity, 96%; and accuracy, 95%). The positive predictive value was 88%, the negative value, 98%. These results support the inclusion of intra-operative frozen-section histology in any protocol for revision arthroplasty for loose components.