Early diagnosis of oral squamous cell carcinoma (OSCC) and precursor lesions is an attractive strategy to decrease patient morbidity and mortality, but presently there are no satisfied diagnostic approaches. This study proposed a metabonomics-based diagnostic approach for OSCC and its precancerous lesions, including oral lichen planus (OLP) and oral leukoplakia (OLK). Saliva samples were collected from patients and healthy donors, and HPLC/MS analysis was performed to acquire metabolic profiles. Diagnostic model was then constructed with hierarchical principal component analysis (HPCA) and discriminate analysis algorithms. The results indicate that metabolic profiling can properly describe the pathologic characteristics of OSCC, OLP and OLK. HPCA combined with kernel fisher discriminant analysis achieved 100% accuracy in diagnosis of test samples, which is superior to direct principal component analysis and other modeling algorithms. The metabonomic approach based on the integral investigation of oral metabolites enables the detection of OSCC and precancerous lesions on noninvasive saliva samples. The proposed approach is noninvasive, efficient and low-cost, and it can be developed as a promising method for population-based screening of cancers and precancers in the oral cavity.