PolyJet 3D printing can be used to fabricate colored physical models of anatomical structures such as skull and heart with realistic appearances. These medical models can be used for surgical simulation and planning of complex operations, as well as anatomy teaching. PolyJet is theoretically capable of producing any color by mixing multiple materials. However, the measured color of a sample printed by PolyJet is often different from the specified color in the printer software. Therefore, it is often difficult to predict the measured color of a sample before printing. This paper reports a study on predictive relationships between measured color and four control factors of PolyJet (i.e., three RGB values of specified color and finish type) by design of experiments and application of multilayer perceptron (MLP) neural network model. Experimental data are collected using a full factorial design of experiments. These data are used to train and test the MLP model using 5-fold cross validation. Then, the prediction performances of the MLP model are compared with a linear regression model and a cubic regression model. The results show that the MLP model is capable of predicting measured color with higher accuracy.