Five standard lactation curve models (incomplete gamma, inverse polynomial, polynomial, exponential and mixed log) were used to predict a typical dairy cow lactation derived as the average daily milk yield of 325 complete first lactation Holstein-Friesian cows in one herd. All but one of these models were fitted to the lactation records of 488 individual cows to determine how well they predict the phenotype of each animal's daily milk production. Accuracy was assessed by the correlation between observed and predicted yield, the magnitude and distribution of the residuals by lactation stage, and the proportion of individual lactations that were well predicted. All models tested predicted the pattern of mean herd lactation well, with little difference in fit. The residuals were mostly non-random, however, with serial correlations indicating biased predictions of yield at certain stages of lactation. The correlation between observed and predicted individual cow's daily milk yields, however, ranged from 0.0 to 0.99 for all models. For some individual lactations, the difference between observed and predicted daily milk yield in certain weeks was as high as 5 kg. The models tested did equally well in predicting typical lactations, which peaked between weeks 6 and 9, and equally poorly in predicting non-typical lactations. The proportion of individual lactations accurately predicted in any group of cows therefore depends mainly on the number whose production follows the typical lactation pattern. (C) 1999 Elsevier Science B.V. All rights reserved.