Breast cancer has never had any good serum tumor markers. Therefore, we developed and evaluated a proteomics approach to searching for new biomarkers and building diagnostic models. SELDI-TOF-MS ProteinChip was used to detect the serum protein patterns of 49 breast cancer patients, 51 patients with benign breast diseases, and 33 healthy women. The diagnostic models were developed and validated using bioinformatics tools such as artificial neural networks and discriminant analysis. In total, four models were built and their sensitivities and specificities were satisfactory. The abilities of these models to diagnose stage I breast cancer were not worse than for stages II-IV (P>0.05). Four candidate biomarkers of breast cancer were found. The high sensitivity and specificity achieved by this method show great potential for the early detection of breast cancer and facilitation of discovering new and improved biomarkers.