In recent years, the K-SVD algorithm has gained more and more attention in the studies of functional magnetic resonance imaging (fMRI) data analysis. In this research, we propose a new method of blind source separation based on non-negative constrained K-SVD (NK-SVD). Firstly, we initialize a dictionary matrix randomly, and use orthogonal matching pursuit (OMP) to obtain a sparse vector matrix. Then, we use NK-SVD to update the dictionary matrix and sparse vector matrix. Furthermore, we solve the dictionary matrix pseudo inverse to obtain the brain functional activation areas by multiplying by the original data. Finally, we apply the proposed method to both simulated data and real fMRI data, where the correspondingly experimental results demonstrate the effectiveness of the proposed one, having better performance in comparison with the conventional algorithms.