Abstract We present a new method to design the controller of Mars capsule atmospheric entry using deep neural networks. Compared to Apollo controller as a baseline, the simulation of neural network controller reproduces the classical Apollo results over a variation of initial position. The deep neural network is only trained with data from Apollo re-entry simulation in Earth model and works in both Earth and Mars environments. It achieves the desired landing accuracy for a Mars capsule. This method works with both linear and nonlinear integration. The results show significant promise in using that approach for future Mars missions.