Instrumental infrared estimation of fat, protein, and lactose in milk: collaborative study.

Thirty-six milks pre-analyzed by accepted standard methods were analyzed with Milko-scan instruments at 6 laboratories. Using pairs comparisons as described by Youden, precision errors for fat were about 0.033% for the first and 0.022% for the second of duplicate tests, indicating a carry-over effect for the first tests. Precision errors for protein were about 0.021% for both tests. Systematic errors for fat were about 0.17% with calibrations based on reference analyses at each laboratory but were reduced to about 0.044% with calibrations by reference analyses from one laboratory. With homogenized milks, systematic errors for fat increased to about 0.07% and were larger at high fat levels. This effect could be minimized by changing the wavelength of maximum transmission for the fat filter. Systematic errors of about 0.067% for protein were reduced to about 0.03% with calibrations based on common reference analyses. Satisfactory calibrations for lactose were obtained with the 2 Milko-scan 203 models with standard errors of estimate of 0.034 and 0.033%. Results for homogenized and unhomogenized milk were different for most instruments, indicating the need to calibrate for them separately.