To tackle denosing problems in hyperspectral remote sensing imagery, a three-dimensional hybrid denoising algorithm in derivative domain was proposed. At first, hyperspectral imagery is transformed into spectral derivative domain where the subtle noise level can be elevated. And then in derivative domain, a wavelet based non-linear threshold denoising method, Bayes-Shrink algorithm, is performed in the two-dimensional spatial domain. In the spectral derivative domain, considering that the noise variance is different from band to band, the spectrum is smoothed using Savitzky-Golay filter instead of wavelet threshold denoising method. Finally, the data smoothed in derivative domain are integrated along the spectral axis and corrected for the accumulated errors brought by spectral integration. The algorithm was tested on airborne visible/infrared imaging spectrometer (AVIRIS) data cubes with signal-to-noise ratio (SNR) of 600:1. Experimental results show that the proposed algorithm can reduce the noise efficiently, and the SNR is improved to more than 2 000:1.