Accurate prediction of convective heat transfer coefficient for supercritical fluids is difficult, owning to the rapid variation of thermal physical properties in pseudo-critical zoon and its complexity. The accuracy and valid range of the existing convective heat transfer correlations for supercritical fluids often cannot be guaranteed at the same time. Drawing lessons from heat transfer to two or multi-phase flow, a new approach to predict the heat transfer to supercritical fluids, called predictive-corrective method, was tried to be introduced. The basis of this approach is the two dimensionless numbers, Bu and Ac , which is able to reflect the strength of buoyancy and flow acceleration effects quantitatively, and a set of validate convective heat transfer correlations for supercritical fluids. Investigation of forced convective heat transfer to carbon dioxide was conducted experimentally, as a result constructing a heat transfer databank with 3696 screened data points, developing 10 convective heat transfer correlations for supercritical carbon dioxide. The evaluation of these new correlations and the predictive-corrective method was made. Result shows that the new introduced predictive-corrective method has the capacity to improve the prediction accuracy of heat transfer to supercritical fluids, and to correct the rough deviation of wall temperature when predicting with a single heat transfer correlation.