Normal mapping can reveal the details of complex geometry object and improve the photorealistic when rendering. However, GPU should compress a great number of normal textures to save the valuable video memory and limited bandwidth just as the common texture compression technology. This paper proposes a Hybrid Adaptive Normal Map Texture Compression Algorithm, which utilizes three useful tactics, which are TSSA, BlockIndexTable and I3DC, to compress the normal map texture according to its different features such as a great many of, few or fewer zero texels existed in the normal map texture. The experimental results and comparisons show that it can reach the compression rate of 5: 1 or more and achieve better image result than others?. What?s more, it will be a good candidate for hardware implementation because of its low cost.