A total of 989,582 test-day records of 160,243 first-parity cows collected from 131 herds of Iranian Holstein dairy cows from 1995 to 2014 by the Animal Breeding and Improvement Centre of Iran, were used to determine the best model for lactation curves of fat to protein ratio (FPR) and somatic cell scores (SCS) in the first lactation. Several mathematical models including the Wood (WD), Wilmink (WL), Rook (RK), Dijkstra (DJ), Narushin-Takma (NT) and Ali and Schaeffer (AS) functions were fitted and compared by four comparison measures; adjusted coefficient of determination (R2adj), residual standard deviation (RSD), Akaike's information criterion (AIC) and DurbinWatson statistic (DW). The NT function was the best model for describing the lactation curves of FPR and SCS in terms of higher R2adj and lower RSD and AIC. The calculated values of DW for FPR and SCS under NT function were 1.99 and 1.86, respectively; implying that the existence of positive autocorrelation between residuals was not important for these traits. The Pearson's correlation coefficients between the actual and predicted records of SCS and FPR values were 0.98 and 0.99 (P<0.01), respectively by fitting NT function.
