The invention provides a BTT missile neural network inversion autopilot FPGA realizing method. The method comprises the steps that a BTT missile state equation model is established; an RBF neural network method is used to compensate the modeling error of the BTT missile state equation; a Backstepping method is used to derive control input, so as to design a BTT missile stability controller; missile model calculating is transplanted into FPGA and realized; when an autopilot runs, a desired attitude signal and an initial state vector are input into the controller; the controller calculates control input, namely rudder deflection, and sends the rudder deflection into a model solver to calculate the new attitude information of a BTT missile; and all state signals are stored into a memory FIFO to form circulation. A simulation result shows that the autopilot has a great control effect and simulation time is greatly shortened, which meets a real-time requirement.