The genetic neural network is established to predict supercritical water steady- state mass flow under natural circulation and the method of mean impact value is used to analyze the sensitivity of parameters. The results show that the predictive values of GNN agree well with the actual values. The errors fall in the limits of ± 10%. Within the parameter range, the steady-state mass flow decreases rapidly with inlet temperature increase. The steady-state mass flow increases with test section height increase or heating zone length, inlet and out...