Electronic gaming machines (EGMs) are one of the most harmful forms of gambling at an individual level. It is unclear whether restriction of EGM functions and accessibility results in meaningful reductions in population-level gambling harm. A natural policy experiment using a large (N = 15,000) national dataset weighted to standard population variables was employed to compare estimates of gambling problems between Australian residents in Western Australia (WA), where EGMs are restricted to one venue and have different structural features, to residents in other Australian jurisdictions where EGMs are widely accessible in casinos, hotels and clubs. Accessibility of other gambling forms is similar across jurisdictions. Gambling participation was higher in WA, but EGM participation was approximately half that of the rest of Australia. Aggregate gambling problems and harm were about one-third lower in WA, and self-reported attribution of harm from EGMs by gamblers and affected others was 2.7× and 4× lower, respectively. Mediation analyses found that less frequent EGM use in WA accounted for the vast majority of the discrepancy in gambling problems (indirect path = -0.055, 95% CI -0.071; -0.038). Moderation analyses found that EGMs are the form most strongly associated with problems, and the strength of this relationship did not differ significantly across jurisdictions. Lower harm from gambling in WA is attributable to restricted accessibility of EGMs, rather than different structural features. There appears to be little transfer of problems to other gambling forms. These results suggest that restricting the accessibility of EGMs substantially reduces gambling harm.