OBJECTIVE:To examine the associations of dietary and supplemental magnesium (Mg) as assessed by a semi-quantitative food frequency questionnaire with cognitive outcomes among ageing women. DESIGN:This work conducts a prospective cohort study of participants enrolled in the Women's Health Initiative Memory Study (WHIMS), which was subsequently extended and named WHIMS-Epidemiology of Cognitive Health. SETTING:Forty clinical centres in the USA. PARTICIPANTS:Postmenopausal women aged 65-79 years without dementia on enrolment. MAIN OUTCOME MEASURES:Physician-adjudicated mild cognitive impairment (MCI) and/or probable dementia (PD). RESULTS:Participants were excluded (n=1006) if they had extreme values of dietary energy intake, had missing or extreme body mass index values, with prevalent MCI/PD at baseline, received only one cognitive assessment or had been followed up for <1 year. During >20 years of follow-up, 765 (11.8%) out of 6473 participants developed MCI/PD. For MCI/PD and MCI, the risks tended to be lower among participants in quintiles Q2-Q5 of Mg consumption compared with those in the lowest quintile. Participants in Q3 had a significantly lower risk of MCI/PD (HR 0.69, 95% CI 0.53 to 0.91) and MCI (HR 0.63, 95% CI 0.45 to 0.87) after multivariate adjustments. No significant association was observed between total Mg intake and PD. The association between total Mg intake, MCI/PD and MCI was non-linear as suggested by the likelihood test. CONCLUSIONS:Total Mg intake between the estimated average requirement and the recommended dietary allowances may associate with a lower risk of MCI/PD and MCI. TRIAL REGISTRATION NUMBER:NCT00685009.