The tropospheric delay is one of the main error sources for radio navigation technologies and other ground- or space-based earth observation systems. In this paper, the spatial and temporal variations of the zenith tropospheric delay (ZTD), especially their dependence on altitude over China region, are analyzed using ECMWF (European Centre for Medium-Range Weather Forecast) pressure-level atmospheric data in 2004 and the ZTD series in 1999-2007 measured at 28 GPS stations from the Crustal Movement Observation Network of China (CMONC). A new tropospheric delay correction model (SHAO) is derived and a regional realization of this model for China region named SHAO-C is established. In SHAO-C model, ZTD is modeled directly by a cosine function together with an initial value and an amplitude at a reference height in each grid, and the variation of ZTD along altitude is fitted with a second-order polynomial. The coefficients of SHAO-C are generated using the meteorology data in China area and given at two degree latitude and longitude interval, featuring regional characteristics in order to facilitate a wide range of navigation and other surveying applications in and around China. Compared with the EGNOS (European Geostationary Navigation Overlay Service) model, which has been used globally and recommended by the European Union Wide Area Augmentation System, the ZTD prediction (in form of spatial and temporal projection) accuracy of the SHAO-C model is significantly improved over China region, especially at stations of higher altitudes. The reasons for the improvement are: (1) the reference altitude of SHAO-C parameters are given at the average height of each grid, and (2) more detailed description of complicated terrain variations in China is incorporated in the model. Therefore, the accumulated error at higher altitude can be reduced considerably. In contrast, the ZTD has to be calculated from the mean sea level with EGNOS and other models. Compared with the direct estimation of ZTD from the 28 GPS stations, the accuracy of the derived ZTD using the SHAO-C model can be improved by 60.5% averagely compared with the EGNOS model. The overall bias and rms are 2.0 and 4.5 cm, respectively, which should be sufficient to satisfy the requirements of most GNSS navigation or positioning applications in terms of the tropospheric delay correction.