The biomarker development in metabolomics aims at discriminating diseased from normal subjects and at creating a predictive model that can be used to diagnose new subjects. From a case study on human hepatocellular carcinoma (HCC), we studied for the first time the potential usefulness of the emerging patterns (EPs) that come from the data mining domain. When applied to a metabolomics data set labeled with two classes (e.g., HCC patients vs healthy subjects), EP mining can capture differentiating combinations of metabolites between the two classes. We observed that the so-called jumping emerging patterns (JEPs), which correspond to the combinations of metabolites that occur in only one of the two classes, achieved better performance than individual biomarkers. Particularly, the implementation of the JEPs in a rules-based diagnostic tool drastically reduced the false positive rate, i.e., the rate of healthy subjects predicted as HCC patients.