Hepatocellular carcinoma (HCC) is one of the most common cancers in the world. The clinical heterogeneity of HCC, and the lack of good diagnostic markers and treatment strategies, has rendered the disease a major challenge. Patients with HCC have a highly variable clinical course, indicating that HCC comprises several biologically distinctive subgroups reflecting a molecular heterogeneity of the tumors. Transforming growth factor β (TGF-β) is known to exhibit tumor stage dependent suppressive (that is, growth inhibition) and oncogenic (that is, invasiveness) properties. Here, we asked if a TGF-β specific gene expression signature could refine the classification and prognostic predictions for HCC patients. Applying a comparative functional genomics approach we demonstrated that a temporal TGF-β gene expression signature established in mouse primary hepatocytes successfully discriminated distinct subgroups of HCC. The TGF-β positive cluster included two novel homogeneous groups of HCC associated with early and late TGF-β signatures. Kaplan-Meier plots and log-rank statistics indicated that the patients with a late TGF-β signature showed significantly (P< 0.005) shortened mean survival time (16.2 ± 5.3 months) compared to the patients with an early (60.7 ± 16.1 months) TGF-β signature. Also, tumors expressing late TGF-β-responsive genes displayed invasive phenotype and increased tumor recurrence. We also showed that the late TGF-β signature accurately predicted liver metastasis and discriminated HCC cell lines by degree of invasiveness. Finally, we established that the TGF-β gene expression signature possessed a predictive value for tumors other than HCC.Conclusion:These data demonstrate the clinical significance of the genes embedded in TGF-β expression signature for the molecular classification of HCC.