1. The present study assesses the accuracy with which mean 24 h breast-milk composition can be estimated if milk samples can only be collected during the daytime. 2. Twenty-five northern Thai mothers, feeding their infants on demand, were studied in their homes for 24 h. All feeds were test-weighed and 0.5 ml pre- and post-feed expressed milk samples taken at each feed. 3. If daytime sampling was restricted to two breast-feeds, it was found that the best estimate of 24 h fat concentration was given by two randomly chosen daytime feeds, predicting 24 h fat concentration with 95% confidence limits of +/- 7.0 g/l (equivalent to 21% of mean 24 h fat concentration). 4. Alternative sampling methods using the mean fat concentration of the feed after 08.00 hours and after 18.00 hours, or the first two feeds after 12.00 hours, predicted fat concentration with 95% confidence limits of +/- 9.7 g/l and +/- 8.9 g/l (28 and 26% of mean 24 h fat concentration) respectively. 5. If well-tolerated by mothers, it would be preferable to sample all daytime feeds, since this reduces the 95% confidence limits to +/- 3.3 g/l, equivalent to 10% of the mean 24 h fat concentration.