In pairwise prediction-error expansion (pairwise PEE) based reversible data hiding (RDH), the correlations among prediction errors are considered and utilized. The obtained performance is better than traditional PEE-based RDH. However, their performance can further be improved. In this letter, a new RDH algorithm based on multiple two-dimensional (2-D) histograms modification is proposed. In our new algorithm, by considering the predetermined pixel pair and its neighbors, a new prediction strategy which can be utilized to predict each element in the pixel pair is proposed. And then, according to the local complexities of the predetermined pixel pairs, multiple 2-D prediction-error sub-histograms are constructed. Moreover, eight new 2-D-maps are designed, and for each kind of 2-D prediction-error sub-histogram, the corresponding optimal 2-D mapping is exhaustively searched to further improve the performance of the proposed algorithm. Extensive experiments demonstrate the superiority of our scheme.