The determination of the minimum miscibility pressure (MMP) between CO 2 and oil holds significant importance in the analysis and modeling of CO 2 miscible flooding processes. Several components in both CO 2 and crude oil could exert complex effects on MMP values. To address this issue in depth, we designed a novel convolutional neural network to investigate the effect of some complex components in CO 2 and oil on MMP in the context of CO 2 enhanced oil recovery. Our results yield valuable insights: (1) employing a multilayer deep convolutional structure, our neural network (NN) model is capable of deeply understanding the relationships of MMP with the contents of various components, extending the type and number of the considered components. Using temperature and the contents of totally 11 components, which do not show obvious cross-correlation, as inputs, the established NN model is demonstrated to attain high prediction precision; (2) temperature has the highest positive correlation with MMP, but the correlation would diminish beyond a certain temperature value. Among the analyzed components, molecular weight of C 7+ and volatile components in oil as well as that of nitrogen, methane, and C 6+ in CO 2 gas show positive correlation with MMP, while mole percentages of intermediate light and heavy hydrocarbons in oil as well as that of hydrogen sulfide, intermediate light hydrocarbon, and C 5 in CO 2 gas exhibit negative correlation. Comparatively, MMP demonstrates heightened sensitivity to variations in molecular weight of C 7+ in oil, while displaying less susceptibility to changes in mole percentage of methane, C 2 −C 4 , and C 5 in gas. The elucidated correlations and sensitivities between MMP and these 12 attributes hold significant potential for future advancements in deriving and refining empirical prediction formulas for MMP.