The CMYK color space was partitioned according to its characteristics. Forward and inverse models of each color space partition conversion are constructed based on RBF neural network, which has the advantages of optimal approximation, global optimization characteristics, and highly nonlinear transformation. A better neural network model was got by sample model training, simulation experiments, and final experimental data analysis and processing. The model has very high precision.