A new approach was developed to retrieve air temperature from remotely sensed data. In a sparse vegetation region, based on the relation between air temperature (Ta) and land surface temperature (TLS), three linear regression equations have been established according to the different grade of elevation. In a dense vegetation region, “Vi-Ts” (vegetation indexsurface temperature) method is used for the estimation of air temperature. Air temperature from this prototype algorithm is validated using NOAAAVHRR data with actual observations of air temperature at China meteorological stations. The error in the sparse vegetation is about 1.5-1.8 ℃(R2=0.89) and the error in the dense vegetation is averagely 1.61 ℃(R2=0.90). The results show that algorithm for estimating air temperature has a high accuracy and can be used in operation.