Recently, Gabor wavelet transformation has been introduced for feature extraction of hyperspectral imagery. Due to the discriminative power of obtained Gabor features, high classification performance has been achieved. However, thousands of Gabor features cause too much burden for onboard computation, limiting the efficiency of the method. In fact, not all features have a positive effect on classification. In this paper, we have proposed a Gabor cube selection-based Multi-task Joint Sparse Representation framework, abbreviated as MT-SG, for hyperspectral imagery classification. Firstly, based on the Fisher discrimination criterion, the most representative Gabor cubes for each class have been picked out. Next, under multi-task joint sparse representation framework, a coefficient vector can be obtained for each test sample with the selected cube features, which can be applied for the following residual-based classification. Experimental results on real hyperspectral data have demonstrated the feasibility and efficiency of the proposed method.