Presence of Mycobacterium avium subsp. paratuberculosis (MAP) in milk for human consumption is a concern due to its possible relationship with Crohn’s disease in humans. Pasteurization effectively reduces the MAP load by four to five logs, but the efficacy depends on the MAP concentration, which depends on the prevalence among contributing herds and individuals. Considerable variation of MAP in bulk tank milk (BTM) and individual cow’s milk (IM) is reported, but factors associated with MAP occurrence in milk at farm level have not been described. This study systematically reviewed published studies aiming at estimating the occurrence of MAP in on-farm BTM and IM by meta-analysis. A total of 692 articles were identified through electronic databases and initially screened using title and abstract. The quality of the 61 potentially relevant articles was assessed using full text and 31 articles were eventually included in the meta-analysis. The apparent prevalence (AP) of MAP in BTM and IM on farm were summarised in relation to strata defined by the test used to identify MAP and the infection status of the herds/animals. There was considerable inconsistency in the reporting, resulting in missing information potentially explaining the dispersion in the estimated AP. The overall AP and 95% confidence intervals based on PCR and culture of MAP were summarised to 0.10 (0.04-0.22) in BTM and 0.20 (0.12-0.32) in IM. Quantifying the MAP load in test-positive milk samples was not possible because very few articles provided quantitative information on individual samples.