Hyperspectral imaging has recently been introduced into face and palmprint recognition and is now drawing much attention of researchers in this area. Compared to simple 2D imaging technology, hyperspectral image can bring much more information. Due to its ablity to jointly explore the spatial-spectral domain, 3D Gabor wavelets have been successfully applied for hyperspectral palmprint recognition. In this approach, a set of 52 three-dimensional Gabor wavelets with different frequencies and orientations were designed and convolved with the cube to extract discriminative information in the joint spatial-spectral domain. However, there is also much redundancy among the hyperpecstral data, which makes the feature extraction computationally expensive. In this paper, we propose to use AP (affinity propagation) based clustering approach to select representative band images from available large data. As the number of bands has been greatly reduced, the feature extraction process can be efficiently speed up. Experimental results on the publicly available HK-PolyU hyperspectral palmprint database show that the proposed approach not only improves the efficiency, but also reduces the EER of 3D Gabor feature based method from 4% to 3.26%.