AIMS: To estimate the treated prevalence of mental illness, substance use disorder (SUD), and dual diagnosis, and the association between dual diagnosis and fatal and non-fatal overdose, among residents of British Columbia, Canada. DESIGN: A retrospective cohort study using linked health, income assistance, corrections, and death records. SETTING: British Columbia (BC), Canada PARTICIPANTS: 921,346 BC residents (455,549 males and 465,797 females) aged 10 and older. MEASUREMENTS: Hospital and primary care administrative data were used to identify a history of mental illness only, SUD only, dual diagnosis, or no history of SUD or mental illness (2010-2014), and overdoses resulting in medical care (2015-2017). We calculated crude incidence rates of non-fatal and fatal overdose by dual diagnosis history. Andersen-Gill and competing risks regression were used to examine the association between dual diagnosis and non-fatal and fatal overdose, respectively, adjusting for age, sex, comorbidities, incarceration history, social assistance, history of prescription opioid and benzodiazepine dispensing, and region of residence. FINDINGS: Of the 921,346 people in the cohort, 176,780 (19.2%), 6147 (0.7%) and 15269 (1.7%) had a history of mental illness only, SUD only, and dual diagnosis, respectively. 4,696 (0.5%) people experienced 688 fatal and 6,938 non-fatal overdoses. In multivariable analyses, mental illness only, SUD only, and dual diagnosis were associated with increased rate of non-fatal (hazard ratio (95%CI): 1.8(1.6-2.1), 9.0(7.0-11.5), 8.7(6.9-10.9), respectively) and fatal overdose (1.6(1.3-2.0), 4.3(2.8-6.5), 4.1(2.8-6.0), respectively) compared with no history. CONCLUSIONS: In a large sample of residents of British Columbia (Canada), approximately one in five people had sought care for a substance use disorder or mental illness in the past five years. The rate of overdose was elevated in people with a mental illness alone, higher again in people with a substance use disorder alone, and highest in people with a dual diagnosis. The adjusted hazard rates were similar for people with substance use disorder only and people with a dual diagnosis.