Epidemiological research often ascertains cancer history via self-reported questionnaires. We assessed the validity of self-reported cancer diagnoses in women born 1921–1926 recruited to the Australian Longitudinal Study in Women's Health (ALSWH) and determined the factors associated with false positive (FP) and false negative (FN) reporting. 4234 ALSWH cohort members were asked at baseline (1996) and in subsequent three-yearly surveys whether they had been diagnosed with specific cancers, including breast, cervical, lung and colorectal. We linked the cohort to the population-based New South Wales Central Cancer Registry (CCR) from 1972 to 2005 to identify registered invasive cancers. We calculated sensitivity, specificity and positive predictive value (PPV) of self-reported cancer diagnoses overall, at baseline (prevalent cancers) and follow-up (incident cancers) using the CCR diagnosis as the ‘gold standard’. We used adjusted logistic regression to examine the determinants of FP and FN reports. Overall sensitivity was 89.2% (95% CI 86.0–91.7%) and exceeded 90% for breast, lung and colorectal cancer at baseline. Overall specificity was 96.9% (95% CI 96.3–97.5%), however, PPV was lower at 66.5% (95% CI 62.7–70.1%). FN reporting of any cancer at baseline was associated with being born overseas. Sensitivity and specificity of self-reported cancer diagnoses in this cohort of older women (aged 70–75 years at baseline) is high but PPV is comparatively lower. Hence, the use of linked data from population-based cancer registries is recommended for studies of cancer epidemiology. Particular attention must also be paid to country of birth in self-reported cancer data, as these findings suggest cancer will be under-reported by this group of women.