Abstract: The feasibility of assay of principal milk constituents by Near -infrare d spectroscopy was studied.The digital fourier filtering pretreatment and wavelength selection by gen etic algorithms in partial least -sq uares(PLS)regressionmodelswerediscussed.Theprincipalmilkconstituentswere assayedwithrootmeansquareerrorofprediction(RMSEP):0.1～0.2g /dl.Therewas goodcorrelationbetweenpredictionvalues and referencevalues withcorrelation coefficienthigherthan0.9.Inaddi tion,theresults werereliablewith coefficientofvariation(CV)ofrepeatabilitybelow0.02.Therefore,Near -infrared spectroscopy could meet practical measurement accuracy as required for principal milk constituents.This study provideda newmethod for nondestructive a nd rapidmeasurement of food.

Key words: Milkconstituents , Near -infraredsp ectroscopy , Partialleast -squares , Digitalfilter , Genetic algorithms