ObjectivesTo evaluate the diagnostic performance of Milan system for reporting salivary gland cytopathology (MSRSGC) in two southern China tertiary cancer centers and investigate the impact of rapid on‐site evaluation (ROSE) on FNAC performance.Materials and MethodsFive hundred and forty‐nine patients who underwent FNAC for salivary lesions with surgical follow‐up from two centers were enrolled in this retrospective cohort study. All slides were recategorized using MSRSGC after consensus on diagnostic criteria for each category. The diagnostic performance of FNAC for salivary lesions was evaluated and compared and the impact of ROSE on FNAC performance was analyzed.ResultsThe distribution of cases per category based on the MSRSGC criteria in the whole series was as followed: ND 49 (8.9%), NN 76 (14.4%), BN 262 (47.7%), AUS 20 (3.6%), SUMP 43 (7.8%), SM 21 (3.8%), M 78 (14.2%). The SUMC series had significantly more ND distributions than JXCH did (16.2% vs. 0, p = .000). Risk of malignancy for each category in the total series was as followed: 42.9% for ND, 9.2% for NN, 3.8% for BN, 30.0% for AUS, 23.3% for SUMP, 81.0% for SM, and 94.9% for M. When ND and AUS/SUMP were excluded, the sensitivity, specificity, PPV, NPV, and accuracy were 84.0%, 97.1%, 89.9%, 95.1%, and 94.0%, respectively; sensitivity, specificity, PPV, NPV, and accuracy were comparable between the two centers.ConclusionsFNAC using MSRSGC provides a good tool in preoperative evaluation for salivary lesions in southern China. ROSE improves its diagnostic performance by reducing the ratio of the ND category.