The diagnosis of Parkinson's disease (PD) still lacks objective diagnostic markers independent of clinical criteria. Cerebrospinal fluid (CSF) samples from 36 PD and 42 age-matched control patients were subjected to inductively coupled plasma-sector field mass spectrometry and a total of 28 different elements were quantified. Different machine learning algorithms were applied to the dataset to identify a discriminating set of elements yielding a novel biomarker signature. Using 19 stably detected elements, the extreme gradient tree boosting model showed the best performance in the discrimination of PD and control patients with high specificity and sensitivity (78.6% and 83.3%, respectively), re-classifying the training data to 100%. The 10 times 10-fold cross-validation yielded a good area under the receiver operating characteristic curve of 0.83. Arsenic, magnesium, and selenium all showed significantly higher mean CSF levels in the PD group compared to the control group (p = 0.01, p = 0.04, and p = 0.03). Reducing the number of elements to a discriminating minimum, we identified an elemental cluster (Se, Fe, As, Ni, Mg, Sr), which most importantly contributed to the sample discrimination. Selenium was identified as the element with the highest impact within this cluster directly followed by iron. After prospective validation, this elemental fingerprint in the CSF could have the potential to be used as independent biomarker for the diagnosis of PD. Next to their value as a biomarker, these data also argue for a prominent role of these highly discriminating six elements in the pathogenesis of PD.