Several studies conducted worldwide report an inverse association between caffeine/coffee consumption and the risk of developing Parkinson's disease (PD). However, heterogeneity and conflicting results between studies preclude a correct estimation of the strength of this association. We conducted a systematic review and meta-analysis of published epidemiological studies to better estimate the effect of caffeine exposure on the incidence of PD. Data sources searched included Medline, LILACS, Scopus, Web of Science and reference lists, up to September 2009. Cohort, case-control and cross-sectional studies were included. Three independent reviewers selected the studies and extracted the data on to standardized forms. Twenty-six studies were included: 7 cohort, 2 nested case-control, 16 case-control, and 1 cross-sectional study. Quantitative data synthesis of the most precise estimates from each study was accomplished through random effects meta-analysis. Heterogeneity was quantified using the I2 statistic. The summary RR for the association between caffeine intake and PD was 0.75 [[95% Confidence Interval (95%CI): 0.68-0.82], with low to moderate heterogeneity (I2= 28.8%). Publication bias for case-control/cross-sectional studies may exist (Egger's test, p=0.053). When considering only the cohort studies, the RR was 0.80 (95%CI: 0.71-90; I2=8.1%). The negative association was weaker when only women were considered (RR=0.86, 95%CI: 0.73-1.02; I2=12.9%). A linear relation was observed between levels of exposure to caffeine and the RR estimates: RR of 0.76 (95%CI: 0.72-0.80; I2= 35.1%) per 300 mg increase in caffeine intake. This study confirm an inverse association between caffeine intake and the risk of PD, which can hardly by explained by bias or uncontrolled confounding.