Abstract: Sub-pixel mapping is an effective method to obtain the distribution of land cover in mixed pixels. This paper proposes a sub-pixel mapping algorithm based on an improved BP neural network with Moran' I, which is a kind of spatial autocorrelation function, to constrain the distribution of sub-pixels to satisfy the concept of spatial dependence better than conventional BP neural network methods. The experimental results indicate that the proposed mapping algorithm outperforms the original BPNN model in terms of both quantitative measurements and visual evaluation.