Background Observational studies have found that both short and long sleep duration are associated with increased risk of metabolic syndrome (MetS). This study aimed to examine the associations of genetically determined sleep durations with MetS and its five components (i.e., central obesity, high blood pressure, dyslipidemia, hypertriglyceridemia, and hyperglycemia) among a group of elderly population.Methods In 335,727 participants of White British from the UK Biobank, linear Mendelian randomization (MR) methods were first employed to examine the causal association of genetically predicted continuous sleep duration with MetS and its each component. Nonlinear MR analyses were performed to determine the nonlinearity of these associations. The causal associations of short and long sleep duration with MetS and its components were further assessed by using genetic variants that associated with short (<= 6 h) and long sleep (>= 9 h) durations.Results Linear MR analyses demonstrated that genetically predicted 1-h longer sleep duration was associated with a 13% lower risk of MetS, a 30% lower risk of central obesity, and a 26% lower risk of hyperglycemia. Non-linear MR analyses provided evidence for non-linear associations of genetically predicted sleep duration with MetS and its five components (all P values < 0.008). Genetically predicted short sleep duration was moderately associated with MetS and its four components, including central obesity, dyslipidemia, hypertriglyceridemia, and hyperglycemia (all P values < 0.002), whereas genetically long sleep duration was not associated with MetS and any of its components.Conclusions Genetically predicted short sleep duration, but not genetically predicted long sleep duration, is a potentially causal risk factor for MetS.