Three-dimensional (3D) wavefront sensing of atmosphere turbulence is a key step to realize multi-conjugate adaptive optics technology. In this article, model tomography, which is one of the most important algorithms for 3D wavefront sensing, is analyzed in theory, the principle limitation of model tomography is proposed. Based on this view the reason for tomography error is discussed, and the simulation results for different errors are produced finally. The analysis shows that part Zernike model basis is used as a new basis of polynomial decomposition in model tomography, which leads to modal coupling error and aliasing error. The un-correlation of part Zernike model basis is the prerequisite to avoid modal aliasing error, but modal coupling error cannot be removed and we can only restrain its influence. Combined with simulation result a method with large field of view (FOV) sensing and small FOV reconstruction is proposed and gives a good performance to control modal coupling error.