The objective of this work was to develop a method to produce reference milks for calibration of mid-infrared (MIR) milk analyzers for milk fatty acid (FA) analysis and validate between laboratory agreement. Recently, MIR PLS models for de novo, mixed origin, preformed milk FA, FA chain length (CL) and FA unsaturation (double bonds/FA), milk estimated blood NEFA have been developed for Delta MIR milk analyzers for use to manage dairy cow feeding and health. Today, 9 commercial instruments are running milk FA analysis in North America using the PLS dairy herd management FA models developed at Cornell and calibrated with reference milk produced at Cornell. The 14-sample calibration set is made once every 4 weeks. Each sample has a reference value in g/100 g of milk for each major milk component and each milk FA, or group of FA measured. Values for each FA, or group of FA, divided by the sum of all FA measured by MIR (not by MIR measured fat test) to provide relative proportions of each that is independent of MIR milk fat concentration. This and the design of the PLS model makes the relative milk FA estimate independent of variation in fat concentration. Nine MIR Delta Instruments were calibrated for de novo, mixed, and preformed FA using the 14-sample calibration sample set 2 weeks before analysis of a set of 8 individual farm validation milks. The 8 validation milks ranged in fat content from 3.4 to 4.8 g/100 g milk in total fat content. For denovo, mixed origin, and preformed FA, FA chain length, and FA unsaturation the mean, mean difference (MD) of instrument minus GLC reference and standard deviation of the difference (SDD) were Mean 0.887, 1.214, 1.550, 14.745, and 0.298, MD −0.016, 0.080, −0.054, −0.06, and 0.00 and 0.028, 0.053, 0.058, 0.033, and 0.010, respectively, on the validation milks. Key instrument factors influence MIR milk fatty acid determination are variation in instrument homogenizer performance and moisture in the optical light path within the instrument.