Considering the dynamic characteristics of Winged-Cone model, the whole channel deviation stability criterion of hypersonic vehicle is reasonably simplified to analysis the coupling influence of multiple parameters. In addition, given the insufficient fitting accuracy of traditional polynomial response surface method, radial basis function neural network, which owns superior comprehensive performance, is selected to achieve aerodynamic data fitting to ensure the fast response and accuracy for sensitivity analysis after comparing the fitting accuracy and robustness of multiple neural networks and PRSM. Finally, the Sobol sensitivity analysis method is used to analyse the sensitivity of angle of attack and Mach number etc. And the deviation stability probability in different angle of attack is calculated by Monte Carlo simulation. The results show that within the range of parameter values, in the criterion, the global sensitivity in ascending order is: angle of attack, rudder deflection angle, Mach number and moment of inertia, meanwhile the probability of stability is relatively high when angle of attack is between the range of 4° to 12°, or 12° to 20°.