We assessed the extent to which using large geographic regions to group ethnicities (ancestries or countries-of-birth) masked intra-regional variation in diabetes risk. We performed a cross-sectional analysis of the 2021 Australian National Census, which included self-reported health data. Ethnicity-specific diabetes prevalence was age/sex-standardised to a reference population of all census respondents 20 years and above. There were 17.5 million adults included in this study. Within four geographical regions, there wastwo-four-fold intra-regional variation in diabetes risk. Diabetes prevalence among people reporting a single East Asianancestry ranged from less than the Australian prevalence (Japanese 4.1%, Thai 6.1%) to twice the Australian prevalence (Filipino 12.2%). Among peoplereporting a single South/Central Asianancestry, diabetes prevalence ranged from 7.1% (Armenian) to 18.9% (Bangladeshi). Among people reporting a single Middle Eastern/North African ancestry, diabetes prevalence values rangedfrom 5.4% (Jewish) to 12.0% (Iraqi). In Oceania, the diabetes prevalence in people of Australian Aboriginal, Fijian, Maori, Samoan and Tongan ancestry was greater than the Australian prevalence(16.7%, 11.9%, 9.9%, 16.0% and 16.6%, respectively versus 6.1%). There wastwo-four-fold variation in diabetes prevalence between populations within four geographical regions. Aggregating ethnicity into large geographic regional groups may incorrectly estimate diabetes risk.