Objectives/Hypothesis Previous studies report high‐accuracy rates for intraoperative frozen sections, but reliability of frozen sections in predicting the ultimate final margin status is unknown. We compared frozen and permanent reads to identify risk factors for overall discrepancies between intraoperative and final margin status. Study Design Retrospective chart review. Methods Pathology reports of 437 surgical resections between 2010 and 2013 were retrospectively reviewed. A total of 253 cases, generating 1,109 individual specimens, met inclusion criteria. Patient demographics, treatment, recurrence, and survival, as well as pathology data pertaining to the specimen, were recorded. Results Frozen read accuracy was 96.7% (83.1% sensitivity, 97.9% specificity) relative to permanent evaluation. However, 4.3% of cases had a final positive margin not detected by frozen section; 17.8% had a close margin not detected by frozen section. In eight of 11 cases with missed positive margins, the involved margin was never sampled intraoperatively. Cases where intraoperative margins were only taken from surrounding tissue had a higher risk of missing a close or positive final margin when compared to cases where some or all margins were taken from the specimen (odds ratio = 5.05, 95% confidence interval 2.31, 11.07, P <0.0001). Disease subsite, risk score, prior radiation, staging, and p16 expression were not significantly associated with the likelihood of missing a close or positive final margin. Conclusion Individual frozen section reads are highly accurate. However, negative intraoperative margins do not guarantee margin‐negative resections. The process of selecting representative margins for intraoperative analysis, specifically the practice of sampling the resection bed, should be refined. Level of Evidence N/A. Laryngoscope, 126:1768–1775, 2016