In this study the Linear functions: Hyperbolic Linear (HLF), Quadratic logarithmic (FQF), and non-linear functions: Parabolic Exponential (PEF), Incomplete Gama (IGF), and Polinomial Inverse (FPI), were employed with the goal to assess what is the function that best fits the curve of lactation average, using data from milk production from twelve herds of buffalo and their Murrah race cross, and reared under intensive systems with once daily milking. We considered details from 1280 lactations with 8839 measurements of milk production. The milk yield determination was conducted on average every 30 days during the period from 1990 to 2001.The lactations employed containing at least four and a maximum of nine measurements if milk production. The criteria used to verify the quality of the adjustment of each function were: the coefficient of determination adjusted (RA2), the test "Lack of Fit (LOF)," the graphic distribution of waste and waste Lag graph. The best ratios were 88.79% for determination FQF followed by the FPI (P> 0.05). All mathematical functions tested underestimated the peak of milk production in Murrah buffaloes, with the quadratic logarithmic function coming closest (adjusted) to peak production.