The prognosis of hepatocellular carcinoma (HCC) remains poor. Vascular invasion, tumour multiplicity and large tumour size are the conventional poor prognostic indicators related to early recurrence. However, it is difficult to predict prognosis of each HCC in the absence of these indicators. The purpose of this study is to predict early recurrence of HCC after radical resection based on whole human gene expression profiling. Microarray analyses were performed in 139 HCC primary tumours. A total of 88 cases lacking the conventional poor prognostic indicators were analysed to establish a molecular prediction system characteristic for early recurrence in 42 training cases with two polarised prognoses, and to test its predictive performance in 46 independent cases (group C). Subsequently, this system was applied to another 51 independent cases with some poor prognostic indicators (group D). The molecular prediction system accurately differentiated HCC cases into poor and good prognoses in both the independent group C (disease-free survival [DFS]: p=0.029, overall survival [OS]: p=0.0043) and independent group D (DFS: p=0.0011, OS, p=0.035). Multivariate Cox regression analysis indicated that the clinical value of molecular prediction system was an independent prognostic factor (p<0.0001, hazard ratio=3.29). Gene expression pattern related to early intrahepatic recurrence inherited in the primary HCC tumour can be useful for the prediction of prognosis.