We performed a retrospective analysis of thirty-three consecutive total hip and knee (twenty-three hip and ten knee) revision arthroplasties during which intraoperative frozen sections were analyzed. Data for the study were collected by means of a review of the charts, radiographic analysis, and evaluation of both frozen and permanent histological sections. The frozen sections, of periprosthetic tissue at the bone-cement interface or the pseudocapsule, were considered positive for active infection if there were more than five polymorphonuclear leukocytes per high-power field in at least five distinct microscopic fields. All patients were available for follow-up, at an average of thirty-six months (range, seventeen to seventy-nine months) after the initial revision operation. The frozen sections from ten patients were positive for infection, and those from twenty-three patients were negative. Comparison of the results of the analyses of the frozen sections (both positive and negative) with those of the analyses of the permanent histological sections of similar tissue showed a correlation of 100 per cent (sensitivity, 1.00; specificity, 1.00; and accuracy, 1.00). Nine patients had positive intraoperative cultures, and all of them had positive frozen sections (sensitivity, 1.00). Of the twenty-four patients who had negative intraoperative cultures, twenty-three had negative frozen sections (specificity, 0.96). Of the nine patients who had positive intraoperative cultures, only two were found to have infection on intraoperative gram-staining. The surgeon's operative assessment regarding the presence of infection, compared with the final pathological diagnosis, demonstrated a sensitivity of 0.70, a specificity of 0.87, and an accuracy of 0.82. All ten patients who had positive frozen sections were managed with excision arthroplasty; six of them subsequently had reimplantation, and the excision was the definitive procedure in the remaining four. One patient who had had a delayed reimplantation had a secondary skin slough and eventually was managed with an arthrodesis of the knee. In the group that had negative frozen sections, eighteen patients had a primary exchange revision arthroplasty and five had a delayed reimplantation. At the time of follow-up, one patient who had had a delayed reimplantation had radiographic loosening of the femoral component and was asymptomatic. One patient who had had a primary exchange arthroplasty was managed with a second revision because of aseptic loosening. There was no clinical recurrence of infection in any patient. The data indicate that analysis of frozen sections of periprosthetic tissue is a reliable predictor of the presence of active infection during revision joint arthroplasty. We recommend its use to differentiate aseptic from septic loosening.