Background: Although abnormalities in gut microbiota are hypothesized to influence the pathogenesis and clinical phenotype of Parkinson’s disease (PD), prospective studies on de novo patients are lacking. Objective: To preliminarily investigate whether gut microbiota in early untreated PD may predict motor and non-motor features progression over a 3-year period. Methods: 16S ribosomal RNA gene amplicons were sequenced on fecal samples of 39 de novo PD patients. Multiple confounders were taken into account, including dietary habits. Motor and non-motor symptoms were assessed using validated scales at baseline and followed-up yearly for 3 years. At last follow-up, a detailed neuropsychological assessment was additionally performed. A general linear model for repeated measurements— adjusted by dopaminergic therapy at follow-up— was used to investigate the relationship between bacterial taxa abundance at baseline (stratified by the median of distribution at baseline) and outcome variables. Results: Twenty-five patients were included (11 refused, 2 lost at follow-up, 1 died). Lower abundance of Roseburia (Firmicutes phylum) at baseline was associated with worse evolution of motor, non-motor and cognitive functions at 3-year follow-up. Similarly, lower abundance of Ruminococcaceae and Actinobacteria at baseline was associated with faster worsening of global cognitive functions. At follow-up, frontal lobe functions were the features most robustly associated with baseline microbial abnormalities. Conclusion: In the present exploratory study on de novo PD, we found an association between abnormal distribution of specific bacterial taxa and the progression of motor and non-motor features over a 3-year period. This proof-of-principle study supports the design of a larger observational study aiming to determine whether these differences survive multiple-comparison correction and define microbiota-specific subgroups suitable for therapeutic targeting.