Based on the statistics and processing of injected gas composition，reservoir temperature，crude oil composition，critical temperature of injected gas，and minimum miscible pressure of 35 CO2 flooding reservoirs at home and abroad，a new GPR-DE model integrating Gaussian process regression（GPR）and differential evolution algorithm（DE）is proposed for predicting the minimum miscibility pressure of a CO2-crude oil system. The accuracy of the GPR-DE model is evaluated with regard to statistical errors and graphical errors. The model results are verified by experimental data and sensitivity analysis and compared with the prediction results of existing models. The results demonstrate that，compared with other models，the GPR-DE model has higher accuracy and wider applicability，with the average absolute relative error of only 2.060% and the standard deviation of only 0.053 2. The GPR-DE model can predict the minimum miscibility pressure of the CO2-crude oil system and other gas and crude oil systems.