Background. Fatty liver index (FLI) is a non-invasive tool used to stratify the risk of non-alcoholic fatty liver disease (NAFLD) in population studies; whether it can be used to exclude or diagnose this disorder is unclear. We conducted a meta-analysis to assess the prevalence of NAFLD in each FLI class and the performance of FLI in detecting NAFLD. Methods. Four databases were searched until January 2021 (CRD42021231367). Original articles included were those reporting the performance of FLI and adopting ultrasound, computed tomography, or magnetic resonance as a reference standard. The numbers of subjects with NAFLD in FLI classes <30, 30–60, and ≥60, and the numbers of subjects classified as true/false positive/negative when adopting 30 and 60 as cut-offs were extracted. A random-effects model was used for pooling data. Results. Ten studies were included, evaluating 27,221 subjects without secondary causes of fatty liver disease. The prevalence of NAFLD in the three FLI classes was 14%, 42%, and 67%. Sensitivity, specificity, positive predictive value, negative predictive value, likelihood ratio for positive results, likelihood ratio for negative results, and diagnostic odds ratio were 81%, 65%, 53%, 84%, 2.3, 0.3, and 7.8 for the lower cut-off and 44%, 90%, 67%, 76%, 4.3, 0.6, and 7.3 for the higher cut-off, respectively. A similar performance was generally found in studies adopting ultrasound versus other imaging modalities. Conclusions. FLI showed an adequate performance in stratifying the risk of NAFLD. However, it showed only weak evidence of a discriminatory performance in excluding or diagnosing this disorder.