Abstract: Study of cloud's automatic detection from the all-sky images is a relative new research field. The current cloud decision algorithm is based on the blue/red ratios for all-sky visible images. However, the aerosol in the atmosphere has different scattering characteristics at different visibilities. So, this static algorithm is not suitable for all kinds of sky conditions. In order to improve the cloud algorithms, we calculated the all-sky radiance at three different wavelengths (450/550/650 nm) under different visibilities and solar angles by using the LIBRADTRAN model. The purpose of this work is to get the distribution character of radiance and radiative ratio (450/650 nm) under different sky conditions. Results show that blue/red radiative ratio will decrease with the decrease of visibility. And thin clouds often demonstrate the same radiative distribution as the ae rosol. The radiative distributions of sky also are different at different solar angles. So, an automatic cloud detection algorithm must include a time-varying factor that is used to adapt to the variation of visibility and solar angle. At present, it is difficult to set up a general function of the visibility and the solar angle because of the complex sky conditions and relationship between them. The efficient and possible method for cloud detection is to set up a radiative dataset of different visibilities and solar angles. A standard database for cloud determination algorithm will be set up.