Some problems existing in current atmospheric dynamic models are proposed, such as vertical discretization, increasing resolution, prediction of the polar regions, discontinuous problems, nonlinear aliasing, computation of the longitudinal finite-difference quotient in high latitudes and so on. In analyzing those problems, possible ways for solving them are suggested and discussed. For example, in discretizing the hydrostatic equation in the vertical, taking lnσas vertical coordinate, instead of σin the case of equal interval, is able to reduce computational errors substantially, and nonlinear aliasing can be neglected only in the case that the factors of the product computation are smooth sufficiently when spherical harmonics are taken as basis functions.