In biometrics, dorsal hand recognition systems are secure, user-friendly, and non-intrusive. Currently, the state-of-the-art in this area uses a single near-infrared spectral band such as 850nm or far-infrared thermography with a range of 8-14µm. This would limit the extraction of only hand veins from a dorsal hand image. However, the dorsal hand contains much more features such as the texture of the skin, pigmented moles, and various chromophores. Therefore, using one spectral band cannot extract all possible dorsal hand features. To resolve this issue, in this paper we propose a novel hyperspectral based dorsal hand recognition system. First, a novel hyperspectral acquisition device is designed to establish a hyperspectral dorsal hand database consisting of 53 spectra. Next, a region of interest was extracted from all spectral dorsal hand images. Then, the partitioned local binary pattern was applied for feature representation extracting both dorsal hand texture and vein features. Finally, the nearest neighborhood classifier was utilized to perform recognition. To validate the proposed system, extensive experiments were conducted on all spectral bands (individually) and the combination of wavelengths of visible light and NIR for both identification and verification using 120 individuals. The highest accuracy for identification (0.998) was achieved using 600nm and 860nm, while for verification the same spectral combo produced the lowest EER of 0.049.