Accuracy of telephone-administered dietary recalls was determined for 159 elderly subjects by comparing the telephone recalls of single midday meals consumed at congregate meal sites with data on actual intake for the meals. Accuracy was determined for kinds of food eaten, size of portions eaten, and content of 15 nutrients. Exact agreement between the two data sources on the kinds of food eaten occurred for only 55% of the observed food items. For 5 of the 10 food groups evaluated, the bias in mean recalled portion-size data was less than 10%. There was evidence of attenuated regression slopes relating recalled to observed portion sizes for 4 of the 10 food groups; adjusted r2 values for the recalled portion sizes ranged from 0.02 to 0.94. Because errors in the recalls of the kinds and amounts eaten tended to go in different directions, the accuracy of the nutrient data was better than that for the food-item or the portion-size data. For 12 of the 15 nutrients evaluated, the bias in mean recalled intake data was less than 10%. Although the adjusted r2 values for the recalled nutrient data ranged from 0.24 to 0.58, there was evidence of attenuated regression slopes relating recalled intake to observed intake for only 5 of the 15 nutrients. We conclude that telephone contact is an acceptable way to obtain short-term dietary recall data from elderly subjects.