ABSTRACT INTRODUCTION: Frozen section is recommended in several situations to: establish the nature of a lesion; establish the presence of a lesion; confirm the presence of a benign lesion; confirm that sufficient tissue is present for diagnosis; establish the grade of the lesion; determine the organ of origin and determine the adequacy of margins. OBJECTIVES: To evaluate the accuracy of frozen section biopsy in multiple organs and analyze possible factors in discrepancy. METHODS: A retrospective study was carried out during a five-year period at a teaching hospital of Recife, Pernambuco, Brazil. The diagnoses of frozen section were compared with results obtained in the permanent section and classified as concordant or discordant. The discordant cases were reviewed by a pathologist and subdivided into false positives and false negatives. Possible reasons for discrepancy were indicated. RESULTS: A total of 1.226 specimens were analyzed, of which 1.181 (96.33%) were concordant and 45 (3.67%) were discordant. After the review of the discordant cases, 39 remained, six (15.4%) were false positives and 33 (84.6%) were false negatives. The tissue that had most false-positive results was mammary sentinel lymph node (3/1.2%), whereas ovarian showed most false negative outcomes with 17 specimens (51.51% of all false negatives). The possible reasons for discrepancy were sampling error, misunderstanding and complexity of the diagnosis. CONCLUSION: The frozen section accuracy of 96.3% found in our study is similar to specialized literature and does not seem to depend on the tissue analyzed.