In order to get more accurate density calculation model of CO2 under different phases which are liquid, gas and supercritical state, the paper applies the relative error to measure the precision of the calculation model based on six typical cubic calculation models and the basic data which is from the U.S. National Institute of Standards and Technology (NIST). It presents a new method based on grayscale image, which is to show the relative error matrix as grayscale image and binary image. It will help to analyze the distribution of relative error for different accuracy. The paper presents a new density calculation model based on composition rules, which improves the accuracy and keep things as simple as possible. The new model clears up relative error at basic data point. The results indicate that the new model is more accurate. The new method in the paper can also be applied to study calculate model of thermo physical parameter for other substances.