A total of 38 patients with frozen section slides who underwent parotidectomy from January 1980 to December 1993 were retrieved from the Department of Pathology. The permanent section slides of the resected specimens were used as the gold standard. The frozen section and permanent section slides were reviewed by two independent pathologists and were labeled either as benign or malignant. If malignant, the tumor grade was determined as either high grade or low grade. Kappas correlation of agreement was used to determine inter-observer variability. Of the 38 slides, 16 were correctly labeled as malignant and 20 as benign. One slide was incorrectly labeled as malignant and 1 as benign. Sensitivity was 94.1 and specificity was 95.2 percent. The prevalence of malignant parotid tumors was 44.7 per cent, with a positive predictive value of 94.1 per cent and a negative predictive value of 95.2 per cent. Kappa statistic for the permanent section was 100 per cent, and 85 per cent for the frozen section. Tumor grading revealed that only 35.2 per cent were labeled correctly as high grade or low grade. This study showed that frozen section could reliably diagnose a malignant parotid tumor but could not reliably determine tumor grade. (Author)