We evaluated the accuracy of 847 consecutive frozen section diagnoses in order to develop a quality control. We also evaluated the time needed to perform them. Frozen sections and final diagnoses agreed in 92.6% and disagreed in 1.7% (14 cases). 5.8% of the cases were deferred. The only case of false-positive frozen sections (0.1%) was due to a pathologist's misinterpretation. False-negative frozen sections were due to sampling errors: in 5 cases, diagnostic tissue was present only in permanent sections of the frozen block and in 8 cases diagnostic tissue was present only in the portion of the specimen not sampled by the frozen section. One hundred and ninety two frozen sections concerned thyroid neoplasms. Thirteen cancers were diagnosed on frozen sections, 2 cancers were considered as benign and 9 cancers had a differed diagnosis. The mean duration to perform the frozen sections was 21 minutes. In conclusion, intra operative frozen section diagnosis is rapid and reliable. Discrepancies are more often false negatives due to sampling errors. Although a high rate of differed diagnosis was observed in thyroid neoplasms, frozen sections remain useful for these lesions. Imprint cytology of thyroid nodules is advisable.