Objective: Using surface-enhanced laser desorption/ionization (SELDI) mass spectrometry to test the salivary protein fingerprint from the patients with thyroid cancer, benign nodular goiter and healthy subjects, to screen out the differentially expressed proteins between each group, and establish a diagnostic model with the help of bioinformatics to provide a novel approach for the early diagnosis of thyroid cancer. Methods: Saliva protein fingerprints of 49 patients with thyroid cancer, 34 patients with nodular goiter and 43 healthy subjects were tested using SEDLI technique. Specific differentially expressed proteins in the saliva of thyroid cancer patient were screened out and a diagnostic model was established with the help of bioinformatics. Results: The total accuracy of the cross validation (test group) for the diagnostic model of thyroid cancer and health control was 81.8% (72/88), sensitivity 88.9% (40/45), specificity 74.4% (32/43); The total accuracy of the cross validation (test group) for the diagnostic model of thyroid cancer and nodular goiter was 89.1% (74/83), sensitivity 87.7%(43/49), specificity 91.1% (31/34). Conclusions: By using SELDI technique, a preliminary diagnostic model for thyroid cancer was established, which may provide a novel approach of high sensitivity and strong specificity for the early diagnosis and pre-operation diagnosis of thyroid cancer. It deserves further investigation and application.