This study selected 90 samples from different brands of red wine. In order to eliminate the impact of spectral curve’s baseline, the first derivatives of all of spectral curves were calculated and the principal component analysis was carried out on the first derivative spectra. The result showed that the contribution rate of the first two principal components was over 80 percent. By the first two principal components, all the red wine samples were obviously divided into two classes. Furthermore a 3-layer artificial neural network predictive model was built with the first four principal components as input variables and 100 percent correct prediction rate was gained. The research showed that the visible-near infrared spectroscopy combined with principal component analysis provides an accurate and reliable new method to rapidly and nondestructively recognize the true or false red wines.