This study put forward a multi-scale evaluation method to reflect classification uncertainty based on hybrid entropy model. In the process of RS images classification, stochastic uncertainty and fuzzy uncertainty are ubiquitous characteristics, and hybrid entropy can be used to measure the total uncertainty of classification caused by both of them. Firstly, we discuss the modeling method based on hybrid entropy theory, and use the statistical data of feature space and fuzzy classifier to obtain modeling parameters. Secondly, pixel hybrid entropy and category hybrid entropy are used as multi-scale indices to evaluate uncertainty of RS images classification. Finally, a remote sensing image of Hubei Province is taking as testing data to verify the above method. Experimental result shows that hybrid entropy can fully reflected stochastic uncertainty and fuzzy uncertainty of classification, and the multi-scale indices are reasonable and effective when compared with the conventional evaluation method which limits to modeling from unitary uncertainty and single scale.