The main objective of this study was to compare the performance of different “quantile regression” (QR) models evaluated at the τth quantile (0.25, 0.50, and 0.75) of somatic cell count (SCC) in Iranian Holstein dairy cows. Mathematical models used in fitting the lactation curve contribute towards better management, physiological and breeding decisions. QR is a flexible tool that allows a specific lactation curve at anyquantile of the trait and can be applied on data with non-normal distributions. Therefore, using QR can be more appropriate, for instance, where the pattern of lactation curve of the trait differs between high and low quantiles. This is more pronounced in trait SCC, where the distribution is not normal and cows with high levels of SCC are suspected to be mastitic cows. Data were collected by the Animal Breeding Center of Iran from 1991 to 2011, comprising 101,147 monthly milk yields of 13,977 cows in 183 herds. Records with DIM <5 and >305 were removed and age at first calving was between 20 and 40 months. An exponential (Wilmink) and a polynomial (Ali & Schaeffer) functions were implemented in the quantile regression. The results showed that all parameters for SCC at the three quantiles Wilmink (a,b and c, parameters) and Ali & Schaeffer function (a, b, c, d and g, parameters) were significantly different from zero (P< 0.01) and not significant, respectively. Parameters b (increasing slope parameter) and c (declining slope parameter) in Wilmink function had increased at the across quantiles and parameter a (SCS level at the beginning of lactation) was not similar. QR with Ali and Schaeffer function fitted the data better than the one with Wilmink function based upon Akaike information criterion and log-likelihood. Among quantiles 0.25th quantile showed best model fit with both functions. QR analysis of SCC, which is a non-normal trait with mixture distribution provides more insight into the management decisions in dairy farms.

Keywords