A three degrees of freedom modeling method for a long-stroke wafer stage in lithography was proposed to solve the X-Y coupling problem of the long-stroke wafer stage and to achieve ultra-precision tracking with micron accuracy. In the modeling method, the rotation angle of X linear motor was considered as one of the controlled objects and the coupling effect on the moving in X direction was involved in the model. Then an adaptive neural network control method was presented based on the proposed model. The Radial Basis Function (RBF) neural network was used to estimate the model information and external nonlinear disturbances real-time online and to reduce the influences of unmodeled dynamics, cogging forces, end effect and friction on the control system. With the theoretical derivation and stability analysis, the convergence of the closed-loop system was guaranteed. Finally, the effectiveness of the modeling and the control method were verified by a S-curve tracking experiment on the actual long-stroke wafer stage of the lithography. The experiment results show that the tracking errors of the X and Y linear motors are less than 3 μm and the rotation angle of X motor is less than 1 μrad. The tracking errors meet the design requirements.