New space gravity missions will provide gravity measurements with unprecedented accuracy and high spatial resolution. To reveal the oceanic and hydrologic signals in monthly time-variable gravity field from the Gravity Recovery and Climate Experiment (GRACE) satellite (Tapley et al., 2004) entails the removal of the atmospheric contribution, which in turn requires a precise knowledge of the atmospheric mass redistribution. We reconstruct the three-dimensional (3-D) variations of air-density from vertical profiles of pressure, temperature, and specific humidity provided by the National Centers for Environmental Prediction (NCEP) atmospheric model of a realistic topography. We compare our results with those from the classical thin layer (2-D) approximation and show that the differences between the complete 3-D and the 2-D computations are often nonnegligible in the presence of the expected GRACE sensitivity up to harmonic degrees of 15--20, corresponding to wavelengths of 2000--2500 km. For actual computation, we recommend the use of the sigma level atmospheric data with special attention to the latitude and altitude dependence of the Earth's gravity. We also examine and conclude the importance of the differences with previous study which assumed a constant surface gravity acceleration without a latitudinal dependence.