We developed a two-step detection approach to map landslides in Chenjiaba area, Beichuan County, Sichuan Province, China after the 2008 Wenchuan earthquake and a strong rainfall four months later. First, the variance information was assessed by image fusion technique. Different from traditional usage of image fusion technique, this paper aims to enhance the interesting features through combing multispectral image with high resolution and panchromatic image with relatively lower resolution. Four fusion technologies (PCA, Brovey, IHS and Wavelet) were tested. All the results contain the spectrum information of both earthquake and rainfall-induced landslides, except the one by wavelet transform based fusion method. The fusion results were assessed by visual inspection and IHS transform based fusion image shows the best performance. Second, fusion image was semi-automatically interpreted. The image interpretation was based on object-oriented analysis which not only the spectral information in pixel-based method was considered, but also the spatial and texture information of the image. Various landscape elements were classified by K Nearest Neighbor algorithm for landslide detection. Accuracy assessment was carried out by comparing those extracted ones with a manually prepared landslide inventory map. Results showed that this approach is capable of mapping different temporal landslides quickly and efficiently.