Spectral computed tomography (CT) has a great superiority in lesion detection, tissue characterization and material decomposition. To further explore its potential clinical applications, in this work, we propose an improved tensor dictionary learning method for sparse-view low-dose spectral CT reconstruction with a constraint of image gradient ℓ0 -norm, which is named as ℓ0 TDL. The ℓ0 TDL method inherits the advantages of tensor dictionary learning (TDL) by employing the similarity of spectral CT images. On the other hand, by introducing the ℓ0 -norm constraint in gradient image domain, the proposed method emphasizes the spatial sparsity to overcome the weakness of TDL on preserving edge information. The results show that the proposed ℓ0 TDL method outperforms other competing methods.