In the present study there were studied and compared linear function (hyperbolic), non linear function (incomplete gama) and segmented polynomials, given by quadratic-quadratic (PSCC) and with three quadratics segmented (PSCCC) models, in the adjustment of average lactation curve of cows with levels of Brown Swiss in their breed composition, with two milkings by day in an semi extensive system. 780 lactations were used, referring to 22318 milk controls. Milk control was done by intervals of 7 days, during 1966 to 2000. The used lactations have a minimum of 22 controls and a maximum of 42. For PSCC, in this production system the curved average reached the "peak" at the 8° week of the lactation, for the PSCCC the "Knots" were considered in 8° and 16° week of lactation. The parameters of each function were considered by active Inter processes through PROC REG of statistical program SAS. The criteria used to verificate the quality of adjustment for each function were: Adjusted Coefficient of Determination (RA2), time of the peak (TP), production in the peak (PP) and remainders distribution graphics. PSCC and PSCCC accompanied the variations in milk production throughout lactation. It was demonstrated that the segmented polynomials present good adjustment for lactation curve.