Neighborhood rough set (NRS) is usually only applicable to small datasets due to the large number of useless and repetitive neighborhood calculations, which severely limits the efficiency of NRS. Many studies improve the efficiency of NRS by narrowing the neighborhood search range down and achieve good performance on small datasets, but they do not perform well on big datasets. To further improve the efficiency on big datasets, we propose a fast attribute reduction method for big datasets based on NRS (FARforBD). In addition, a theorem is also represented to prove the correctness and effectiveness of the proposed method. In FARforBD, we further reduce the neighborhood search range to a neighborhood without any positive region samples. This method greatly reduces many useless neighborhood calculations. The comparison experiments on big datasets show the effectiveness and efficiency of FARforBD.