Early diagnosis of hepatocellular carcinoma (HCC) remains challenging to date. Characteristic metabolic deregulations of HCC may enable novel biomarkers discovery for early diagnosis. A capillary electrophoresis-time of flight mass spectrometry (CE-TOF/MS)-based metabolomics approach was performed to discover and validate potential biomarkers for HCC from the diethylnitrosamine-induced rat hepatocarcinogenesis model to human subjects. Time series sera from the animal model were evaluated using multivariate and univariate analyses to reveal dynamic metabolic changes. Two independent human cohorts (populations I and II) containing 122 human serum specimens were enrolled for validations. A novel biomarker pattern of ratio creatine/betaine which reflects the balance of methylation was identified. This biomarker pattern achieved effective classification of pre-HCC and HCC stages in animal model. It was still effective in the diagnosis of HCC from high-risk patients with cirrhotic nodules, achieving AUC values of 0.865 and 0.905 for two validation cohorts, respectively. The diagnosis of small HCC from cirrhosis with an AUC of 0.928 highlighted the potential for early diagnosis. This ratio biomarker can also improve the diagnostic performance of α-fetoprotein (AFP). This study demonstrates the efficacy of present strategy for biomarker discovery, and the potential of metabolomics approach to provide novel insights for disease study.