This study assesses the validity and reproducibility of a 145-item self-administered food frequency questionnaire (FFQ) in a representative older population aged 63 to 80. Semi-quantitative FFQs were completed by 89% of 3,654 residents attending a community-based eye study in Sydney, Australia. The FFQ's validity was assessed against three, four-day weighed food records (WFRs) completed four months apart by 79 people. A further 152 subjects completed a repeat FFQ about a year after the baseline FFQ, of whom 131 completed a second repeat FFQ about six weeks later. Both short and long-term reproducibility of the FFQ were assessed using data from these subjects. Comparison of the FFQ with the average of the three, four-day weighed food records resulted in energy-adjusted Spearman correlations above 0.5 for most of the nutrients. The proportion of subjects correctly classified to within one quintile category for each nutrient intake ranged from 57% for zinc to 82% for vitamin C. with most nutrients correctly classified within one quintile for about 70% of subjects. Quadratic weighted kappas were reasonable, between 0.3 and 0.5 for most nutrients. The FFQ was highly reproducible in the short term, with correlations for most nutrients about 0.70 to 0.80 and acceptably reproducible in the longer term, with correlations mostly 0.60 to 0.70. The results verify that it is possible to use relatively simple, but comprehensive, self-administered FFQs to study nutrient exposures in large-scale epidemiological studies of the elderly and to expect reasonably high FFQ response rates.