Abstract: The NIR spectrums of health wine was collected by on-line monitoring system,spectrums were preprocessed by first derivative( FD) and norris derivative filt( ND),the NIR model about alcohol content was built by partial least square( PLS).The RMSEC of the model was 0.737,the corr.coeff was 0.9189.The RMSEP of the model was0.788,the corr.coeff was 0.9425.According to the experiment data,the relative difference of alcoholic degree numbers between calculated by near-infrared and standard technique was mainly in the range of ± 2%,which meant this method was able to meet the demand of online testing alcoholic degree during the productive process.