Accurate estimate of direct solar radiation is a prerequisite for the design of concentrated solar power thermal systems; however, direct radiation observations are very sparse compared to observations of routine meteorological variables. Therefore, this study developed a method for estimating daily direct radiation from routine observations, based mainly on a clear-sky physical model to compute clear-sky direct radiation and a parameterized formula to compute cloud transmittance from sunshine duration data. The parameters of the cloud transmittance formula were calibrated with high-quality radiation data measured in the United States. The validation results at independent stations in the United States were comparable to those at the calibration stations, indicating that the method is suitable for various climatic and geographic conditions. We applied this method to compile a long-term, high-density data set of daily direct radiation over China based on observations from 2,473 routine weather stations of the China Meteorological Administration (CMA) network. Although the root-mean-square error of the data set from the 20 CMA radiation stations at which the direct solar radiations have been observed since 1993 is about 27.3Wm(-2), the accuracy of the data set for CMA routine weather stations with less severe air pollution should be better. The newly developed data set in this study, which is available upon request, will contribute to the development of concentrated solar power thermal systems throughout China.