Longitudinal measurements of breast milk output were made using a 2H2O tracer technique over 24 months in 23 rural Papua New Guinean women. A small (n = 14) cross-sectional sample of women from a different area were also studied. Mean outputs of the longitudinal sample rose from 601 +/- 100 g/d at 1 month to a maximum of 901 +/- 211 g/d at 9 months and fell to 501 +/- 188 g/d at 24 months, values which are considerably higher than previously reported from Papua New Guinea. This is considered to be primarily due to the suitability of the method of measurement. The late and complementary nature of the introduction of additional foods to the infant diet is considered to be responsible for the maintenance of these high outputs over periods of up to 24 months in some cases. Multiple regression analysis performed on all data points up to 6 months or 6 kg infant weight showed that infant weight accounted for 55.8 per cent of the total variation in milk output. The additional effect of infant age, given weight, was not significant (t = 1.37). In cross-cultural comparisons infant size was also shown to be a more important determinant of milk output than infant age. These comparisons were made using the rate constant Fbm/Vb, the basic measurement of the estimation of milk output by the tracer method.