The work aims to establish a gravure spot-color matching model based on GRNN neural network in order to realize computer color matching, in view of the current status of spot-color matching. Color space was divided according to Munsell color cycle. Training samples were obtained by preparing spot-color samples which were uniformly distributed in the Munsell color spectrum. By analyzing the advantages of general regression neural network and the characteristics of gravure spot-color matching, the gravure spot-color matching model was constructed upon attempt based on GRNN neural network. Matalb was used for simulation training, and then the value of smoothing factor SPREAD was determined by means of MSE function. Finally, the accuracy of the color matching model was tested by the color difference between target color and matching color of tested samples. Through network simulation, the MSE value of tested samples was determined to be the minimum when SPREAD was 6.4. A color matching model was thus determined. The average color difference between target color and matched color of 40 groups of tested samples was 2.45, and the color difference of 97.5% samples was less than 6. The gravure spot-color matching model based on GRNN neural network has higher precision and can be used in computer color matching.