The role of spousal education on dementia risk and how it may differ by gender or race/ethnicity is unknown. This study examines the association between one’s own education separate from and in conjunction with spousal education and risk of dementia. Cohort. Kaiser Permanente Northern California (KPNC), an integrated health care delivery system. 8835 members of KPNC who were aged 40-55, married and reported own and spousal education in 1964-1973. Dementia cases were identified through medical records from 1 January 1996 to 30 September 2017. Own and spousal education was self-reported in 1964-1973 and each was classified as four indicator variables (≤high school, trade school/some college, college degree and postgraduate) and as ≥college degree versus college degree. Age as timescale weighted Cox proportional hazard models adjusted for demographics and health indicators evaluated associations between participant education, spousal education and dementia risk overall and by gender and race/ethnicity. The cohort was 37% non-white, 46% men and 30% were diagnosed with dementia during follow-up from 1996 to 2017 (mean follow-up=12.7 years). Greater participant education was associated with lower dementia risk independent of spousal education, demographics and health indicators. Greater spousal education was associated with lower dementia adjusting for demographics but became non-significant after further adjustment for participant education. The same pattern was seen for spousal education ≥college degree (not adjusting for participant education HRspousal education≥college degree=0.83 (95% CI: 0.76 to 0.90); adjusting for participant education HRspousal education≥college degree=0.92 (95% CI: 0.83 to 1.01)). These associations did not vary by gender or race/ethnicity. In a large diverse cohort, we found that higher levels of participant’s own education were associated with lower dementia risk regardless of spousal education. An inverse association between spousal education and dementia risk was also present, however, the effects became non-significant after adjusting for participant education.