[en] Calibration equations were developed to quantify fatty acids (FA) in milk using mid-infrared spectrometry. Belgian equations used showed a good ability to predict the contents of saturated (SAT) monounsaturated FA (MONO) Cross-validation coefficient of determination were 99.54% and 98.23% for SAT and MONO. respectively. Some outliers can appear and have impact on the estimation of breeding values. The aim of this study was to develop methods of outlier detection. The first method determined limitations based on individual examinations of extreme values, The second model was based on the residue estimated from random regression test-day models. The dataset contained 58,443 test-day records collected from 16,470 first parity Luxembourg Holstein cows in 699 herds. Models included as fixed effects: herd*test date, days in milk, and age. Random effects were regressed using Legendre polynomials of order 2 and were: herd*year of calving, permanent environment, and animal effect. The residual effect was assumed to be constant through the lactation. Detection by a random regression approach gave better results. The replacement of outlier values by the conresponded expected values was the better approach to calculate reliable breeding values.