This paper introduces a new statistical method to estimate global and direct solar radiation values at hourly intervals by the pixel values of the geostational meteorological satellites images. The feature of the model is to utilize the three channels, i.e., visual, infrared, and water vapor channels of satellite images while other satellite-derived models use only the visual channel. The surface observed hourly cumulative global radiation and direct horizontal radiation at the Fukuoka Meteorological Observatory and corresponding pixel values of the satellite images for a year are screened and classified into three data sets according to the visual channel as an index of the weather, namely, clear, fine and cloudy. The multiple linear regression analyses for the global radiation or the direct horizontal radiation processes three data sets respectively. The result of regression analysis showed the reasonable accuracy of the model.