Metabonomic approaches based on spectroscopic data are in their infancy in biomedicine. A key challenge in clinical metabonomics is uncovering and understanding the relations between the multidimensional spectroscopic data and the clinical measures currently used for disease risk assessment and diagnostics. A novel Bayesian approach for revealing clinically relevant signals is presented here for a real 1H NMR metabonomics data set. The results are not only mathematically superior but also biochemically fully coherent.