Accurate prediction of supercritical CO 2 (scCO 2 ) heat transfer is important for heat exchanger design and safe operation of scCO 2 power cycles. The main prediction method is empirical correlation. This paper demonstrates an alternative way by artificial neural networks (ANN) model with two hidden layers. To assess widely cited correlations and newly developed ANN model, scCO 2 heat transfer experiment in vertical tube with pressure up to 20.8 MPa was performed to extend experiment database, which includes 2674 runs. Compared with empirical correlations, the ANN model is promising for following advantages: (1) ANN model has much better prediction accuracy. The mean relative error, mean absolute relative error and the root-mean-square relative error between predicted and measured wall temperatures are e A =0.38%, e R =4.88% and e S =7.29%, respectively. (2) ANN model performs faster computation speed. (3) ANN model can accurately and speedily predict scCO 2 heat transfer performance for both normal heat transfer and heat transfer deterioration modes. The trained ANN program is provided with this paper, which is a useful tool and can be directly applied in engineering of scCO 2 heat transfer.