OBJECTIVE: To determine the accuracy of salivary gland ultrasonography (SGUS) for diagnosing primary Sjögren's syndrome (SS) and to suggest modifications of the American-European Consensus Group (AECG) classification criteria. METHODS: We conducted a cross-sectional study in a prospective cohort of patients with suspected primary SS that was established between 2006 and 2011. The echostructure of the bilateral parotid and submandibular glands was graded from 0 to 4, and the gland size was measured; blood flow to the parotid gland was assessed using Doppler waveform analysis. The reference standard was a clinical diagnosis of primary SS as determined by a group of experts blinded to the results of SGUS. Receiver operating characteristic (ROC) curve analysis was performed to compare the diagnostic value of the 0-4-point echostructure grade for each of the 4 major salivary glands, the sum of the grades for the 4 glands, and the highest grade among the 4 glands. RESULTS: Of the 158 patients in the study, 78 had a diagnosis of primary SS according to the experts, including 61 patients (78.2%) who met the AECG criteria. Doppler waveform analysis and gland size measurement showed poor diagnostic performance. The results of ROC curve analysis showed that the highest grade among the 4 glands provided the best diagnostic value. The optimal grade cutoff was 2 (62.8% sensitivity and 95.0% specificity). A weighted score was constructed using scores for the 5 variables selected by logistic regression analysis, as follows: (salivary flow × 1.5) + (Schirmer's test × 1.5) + (salivary gland biopsy × 3) + (SSA/SSB × 4.5) + (SGUS × 2). According to ROC curve analysis, a score of ≥5 of 12.5 had 85.7% sensitivity and 94.9% specificity, compared with 77.9% sensitivity and 98.7% specificity for the AECG criteria. The addition of SGUS to the AECG criteria increased sensitivity to 87.0% but did not change specificity. CONCLUSION: Modifications of the AECG criteria, including the addition of a SGUS score, notably improved diagnostic performance.