Mediation analysis is a common statistical method used to investigate mechanisms of health exposure and interventions. The reporting quality of mediation studies used in randomised controlled trials has been considered heterogeneous and incomplete. The reporting quality of mediation analysis in observational studies is unknown. We conducted a systematic review to describe the reporting standards of recently published observational studies that used mediation analysis to understand the mechanism of health exposures. We searched for studies published between June 2017 and June 2019 indexed in EMBASE, MEDLINE and PsycINFO. Two reviewers screened articles and selected a random sample of 50 eligible studies for inclusion. We included studies across 13 healthcare fields and ten different health conditions. Most studies (74%) collected data on healthy individuals to assess their risk of developing a health disorder. Psychosocial and behavioural factors (self-control, self-esteem, alcohol consumption, pain) were the most prevalent exposures (n = 30, 60%), outcomes (n = 23, 46%) and mediators (n = 29, 58%). Most studies used a cross-sectional design (64%, n = 32), and a few studies reported sample size calculations (4%, n = 8). In 20% (n = 10) of the studies, adjustment for confounders was reported. Only 10% (n = 5) of studies reported the assumptions underlying the mediation analysis, and 14% (n = 7) of studies conducted some sensitivity analysis to assess the degree which unmeasured confounders would affect the estimate of the mediation effect. Mediation analysis is a common method used to investigate mechanisms in prevention research. The reporting of mediation analysis in observational studies is incomplete and may impact reproducibility, evidence synthesis and implementation.