The work aims to propose a method of color characterization of a BP neural network with genetic algorithm optimization in the subspace, so as to improve the reproduction accuracy of the color image in digital printing. The basic principle of BP neural network with genetic algorithm (GA) optimization was introduced. A color characterization model established in L*a*b* color subspace was designed, and a GA-BP neural network model training experiment was conducted for 1,000 color samples. The training experiment finally fitted the nonlinear relationship between the L*a*b* chromaticity value of the printed color sample and the inputted RGB driving value of the printed image. The color characterization of 125 test color samples were predicted by the proposed method. The predication results showed that, more than 90% of the color differences were distributed within 2.0 and the spectral RMSE was within 0.02. Compared with the BP neural network without genetic algorithm optimization, the prediction accuracy of the proposed method is obviously improved, which can achieve high reproduction accuracy of color image in digital inkjet printing.