Day-to-day variation within an individual's dietary intake may obscure associations between diet and disease. We calculated the extent of misclassification of subjects by dietary polyunsaturated to saturated fatty acid (P/S) ratio for 1, 3, 7 and 19 survey days per person. In 59 young women diet was estimated 19 times over a period of two and a half years by a 24-hour recall method. The dietary P/S ratio was compared with the P/S ratio in adipose tissue biopsies, which was used as the reference. The proportion of subjects properly classified increased with the number of survey days. Between three and seven recalls appeared to be most cost effective. The effect of misclassification on the observed odds ratio in a case-control study was calculated for a hypothetical distribution of exposure for cases and controls.