In order to select the model of the lactation curve that has the best adjustment for milk production records of Holstein cows in the Minas Gerais State, a total of 157,932 test day of 17,150 lactations from 305,506 test day of 56,422 lactations, from 1990 to 1996, provided by the milk recording service of the ACGHMG (Holstein Breed Association of Minas Gerais), were used. The data were adjusted for: region, management level, year -season of calving, COPI (parity and age of calving), genetic group, number of tests, interactions and linear and quadratic effects of lactation period. In order to compare results obtained using quadratic logarithmic, incomplete gamma, multiple regression and multiphasic (monophasic and diphasic) functions methods. The following statistics were used: lowest standard error, adjusted R², first order auto-correlation and Durbin-Watson. It was concluded that the diphasic function was the best for predict milk yield in selection programs being possible to estimate the 305-day milk yield with this function.
adjustment of models; incomplete gamma; multiphasic; quadratic logarithmic; projection of lactation
