Abstract: Hyperspectral image processing had attracted high attention in remote sensing fields. One of the main issues was to address the problem of huge data and hard transmission via sampling and reconstruction. Compressed sensing theory was investigated in this paper for band reconstruction. Based on compressed sensing theory, original signal could be reconstructed efficiently without satisfying the Nyquist-Shannon criterion. Adjacent spectral bands of hyperspectral images were highly correlated, resulting in strong sparse representation. This significant property made it possible to obtain the whole spectrum information from limited bands of original hyperspectral data via compressed sensing theory. Experimental results demonstrate the feasibility and reliability of applying compressed sensing theory for sampling and reconstruction on bands of hyperspectral images. The proposed band reconstruction method can perform high correlation coefficients and low relative errors between a pair of reconstructed and original hyperspectral bands. Simultaneously, high levels of reconstruction efficiency are achieved, and reconstructed spectral curve is in accordance with original data as well.