BACKGROUND: Benzodiazepine use is highly prevalent in elderly and late middle-aged populations and may be associated with an increased risk of dementia. Observational studies have suggested that benzodiazepine use may increase the risk of dementia, however there have been significant concerns regarding protopathic bias in these studies, precluding conclusive findings. OBJECTIVE: The aim of our study was to investigate the risk of dementia associated with the use of benzodiazepines in elderly patients, after controlling for protopathic bias. METHODS: We identified observational studies with more than 50 cases, adequate assessment of benzodiazepine exposure, and reliable dementia diagnosis ascertainment, from the MEDLINE, PubMed, EMBASE, CINAHL, LILACS and CENTRAL electronic databases through to 5 June 2018, with no language limits. The association of any current or former use of short- or long-acting benzodiazepines with incident dementia was analysed. A subgroup analysis was performed by the introduction of lag time to assess the effect of protopathic bias. We also performed analyses considering the effect of higher benzodiazepine cumulative doses and adjustment for psychiatric covariates. Study quality was investigated using the Newcastle-Ottawa Scale. RESULTS: We identified 15 studies reported in 14 articles, involving 159,090 cases. Ever use of benzodiazepines was associated with a significantly increased risk of dementia [odds ratio (OR) 1.39, 95% confidence interval (CI) 1.21-1.59]. Those studies that implemented the longest lag times of ≥ 5 years, and hence most likely to overcome protopathic bias, found a risk estimate that was marginally attenuated, but still significant (OR 1.30, 95% CI 1.14-1.48). Long-acting benzodiazepines were associated with a marginally higher magnitude risk (OR 1.21, 95% CI 0.99-1.49) than short-acting benzodiazepines (OR 1.13, 95% CI 1.02-1.26), although the former failed to reach statistical significance (p = 0.059). CONCLUSIONS: Our findings indicate that the association between benzodiazepine use and dementia incidence is not purely an artefact due to protopathic bias. Reduction of inappropriate benzodiazepine prescription is likely to attenuate dementia risk.