Land cover products are the important background for researches such as climate change, material and energy cycle, eco-environment evaluation, land surface process modeling, eco-parameters inversion, and so on. The evergreen and deciduous characteristic of forests is one of the most important attributes for land cover products. But it is still a challenge how to efficiently distinguish the evergreen and deciduous characteristic of forests using remote sensing technology at regional or global scales. Especially in the mountainous areas, due to the abundant biodiversity and heterogeneous landscape patterns caused by unique climate, eco-environmental conditions, and long-term and persistent human disturbances, it is usually more difficult to automatically identify the evergreen and deciduous characteristic of forests than other regions. This paper proposed a simple, practical and automatic method to identify the evergreen and deciduous characteristic of forests in mountainous areas. The NDVI (Normalized Difference Vegetation Index) which is the best indicator to represent the growth state of vegetation was selected as an index, and the evergreen needleleaf forest was then selected as reference sample in the proposed method. Firstly, a preliminary map of forest types must be produced by multi-source and multi-temporal remote sensing images through the object-oriented classification method. The frequency histogram of the NDVI_D (the differences of NDVI) between growing season and non-growing season of needleleaf forests obtained from the preliminary map was used to choose the threshold value. The evergreen and deciduous characteristic of forests were accurately distinguished by threshold rules at last. For the areas covered with clouds, seasonal snows and shadows, the evergreen and deciduous characteristics of forests were replaced by the characteristics of its surrounding forests. Choosing the proper threshold values and building the distinguished rules are the cores of this method. This paper took Mt. Gongga as study area, the Landsat TM images, multi-spectral HJ (Chinese small constellation of environmental and disaster mitigation) CCD images and the combined Landsat TM and HJ images to respectively validate the effectiveness of this method. The validation results showed that the proposed method in this paper could effectively identify evergreen and deciduous characteristic of forests in mountainous areas. The total accuracy of identification results was 93.87% and the Kappa coefficient was 0.87. The time phase of remote sensing images, cloud contamination, seasonal snows cover, and shadows cast by mountains and clouds are the major factors affecting the identification accuracy. To use the proposed method, both the time phase and quality of remote sensing images need to be considered. Meanwhile, the remote sensing images covered with clouds or seasonal snows need try to be avoided. This method can be used not only in mountainous areas, but also in the plain or hill regions. However, it is still necessary to choose proper land cover type like the evergreen needleleaf forests in mountainous areas as the reference sample when it is applied in plain and hill regions. The reference sample must geographically widely distributes in the whole area and has small spectral changes in the entire growth cycle. This method is expected suitable for automatic identification the evergreen and deciduous characteristic of forests at large area and had been successfully applied in the "National Ecological Environment Decade of Change (2000-2010)" specific project of MEP&CAS to map the land covers in Southwestern China.