A prospective study was performed to determine the reliability of analysis of intraoperative frozen sections for the identification of infection during 175 consecutive revision total joint arthroplasties (142 hip and thirty-three knee). The mean interval between the primary and the revision arthroplasty was 7.3 years (range, three months to twenty-three years). To reduce selections bias, tissue was obtained for frozen sections during all revisions in patients who did not have active drainage from the wound or a sinus tract. Of the 175 patients, twenty-three had at least five polymorphonuclear leukocytes per high-power field on analysis of the frozen sections and were considered to have an infection. Of these twenty-three, five had five to nine polymorphonuclear leukocytes per high-power field and eighteen had at least ten polymorphonuclear leukocytes per high-power field. The frozen sections for the remaining 152 patients were considered negative. On the basis of cultures of specimens obtained at the time of the revision operation, nineteen of the 175 patients were considered to have an infection. Of the 152 patients who had negative frozen sections, three were considered to have an infection on the basis of the results of the final cultures. Of the twenty-three patients who had positive frozen sections, sixteen were considered to have an infection on the basis of the results of the final cultures; all sixteen had frozen sections that had demonstrated at least ten polymorphonuclear leukocytes per high-power field. The sensitivity and specificity of the frozen sections were similar regardless of whether an index of five or ten polymorphonuclear leukocytes per high-power field was used. Analysis of the frozen sections had a sensitivity of 84 per cent for both indices, whereas the specificity was 96 per cent when the index was five polymorphonuclear leukocytes and 99 per cent when it was ten polymorphonuclear leukocytes. However, the positive predictive value of the frozen sections increased significantly (p < 0.05), from 70 to 89 per cent, when the index increased from five to ten polymorphonuclear leukocytes per high-power field. The negative predictive value of the frozen sections was 98 per cent for both indices. The current study suggests that it is valuable to obtain tissue for intraoperative frozen sections during revision hip and knee arthroplasty. At least ten polymorphonuclear leukocytes per high-power field was predictive of infection, while five to nine polymorphonuclear leukocytes per high-power field was not necessarily consistent with infection. Less than five polymorphonuclear leukocytes per high-power field reliably indicated the absence of infection.