Because the existing reversible information hiding algorithm has low embedding capacity and is not suitable for multispectral images, a multilevel reversible information hiding method for multispectral images based on histogram shifting of prediction error and band re-ordering was proposed. For each hiding level, firstly, the bands were re-ordered by the proposed approach, then, an adaptive predictor was used to exploit the spatial correlation and spectral correlation of multispectral image, finally, the secret information were embedded by a histogram shifting mechanism based on prediction error. The experimental results on multispectral images acquired by Landsat show that the proposed method has better visual quality and higher hiding capacity than the representative methods.