Cloud classification using satellite images has been done for many years, while the study of ground-based cloud classification is still underway. This seems to be difficult because of their local characters, complex shapes and quickly changes. In resent years, sky condition classification was suggested instead of cloud classification. In this research, a method using Fuzzy Uncertainty Texture Spectrum to classify sky conditions is proposed. The method is firstly applied to 200 simplex sky conditions images obtained from the whole sky infrared cloud measuring system in Nanjing, China. The classification accuracy rates of stratiform, cumuliform, waveform, cirriform clouds and clear sky compared with human observations are 100%, 82.5%, 80%, 82.5%, and 100% respectively, the average accuracy rate is 89%. While the method is applied to 217 complex sky conditions, the accuracy rates are reduced to 78.6%, 61.8%, 66.7%, 57.7%, and 82.0% respectively, and the average accuracy rate is only 69.6%. It shows that the complex sky conditions are still difficult to be classified only by using the texture-based feature extraction methods.