This study assesses if nuclear magnetic resonance (NMR) metabonomics can discriminate between Parkinson's disease (PD) patients and control subjects, and consequently identify metabolic markers for the disease. One-dimensional H-1 NMR spectroscopy was used for quantitative analysis of metabolites in the cerebrospinal fluid (CSF) from 10 PD patients and 10 control individuals, together with uni- and multivariate statistical analysis to discriminate between the groups and to identify significantly altered metabolite concentrations. In total 60 metabolites were identified and of those 38 were quantified in all CSF samples. An overall lowering of metabolite content was observed in PD patients compared to control subjects (fold change of 0.85 +/- 0.30). Multivariate statistics reveal significant changes (vertical bar w*vertical bar>0.2) among nine metabolites (alanine, creatinine, dimethylamine, glucose, lactate, mannose, phenylalanine, 3-hydroxyisobutyric acid and 3-hydroxyisovaleric acid). Three of these (alanine, creatinine and mannose) are identified as significantly changed also by univariate statistics (p < 0.00132, Bonferroni corrected). Panels with all or a selected set of these metabolites were successfully used for discriminating between the two groups. In conclusion, NMR metabonomics can readily determine metabolite concentrations in CSF, identify putative biomarkers that distinguish between the PD patients and control subjects, and thus potentially become a tool for diagnostic purposes.