In this paper,CA transformation was introduced, instead of PCA transformation, and integrated with Kohonen Self Organization Feature Map (SOFM) ANN for Landsat ETM+ data classification. The methodology mainly included three steps as follows: First, the non-Lambertian Minnaert topographic correction algorithm was used to remove the topographic effects of the ETM+ image after atmospheric correction from the test site. Second, the ETM+ image after topographic correction was transformed using the CA algorithm. Then, the SOFM ANN analysis was applied to the CA first two components selected to perform Land Use/Land Cover (LULC) classification. And the results suggested that the proposed approach is more effective for LULC classification of ETM+ image than the approach based on PCA for the test site, and also showed that topographic correction is necessary for Landsat ETM+ images from rugged terrain and helpful to improve the classification accuracy.