Abstract: Mixed pixels will always be present in remote sensing images, soft classification techniques have been developed to estimate the class composition of mixed pixels, and the accuracy of land cover mapping has been improved, but their output provides no indication of how these classes are distributed spatially in pixels. Sub-pixel mapping is a technique to produce the land cover map at sub-pixel spatial resolution from the land cover proportion images obtained by soft classification methods. In this technique, pixels are divided into sub-pixels, and these fraction values can be assigned to sub-pixels, based on the assumption of spatial dependence. Sub-pixel mapping can represent the land cover class fractions, so it can provide better spatial representation of land cover. A new algorithm is presented for sub-pixel mapping, the algorithm is based on the scale of sub-pixels spatial attraction models, which can express the spatial dependence well. And in this algorithm, taking into account the interaction of pixels between themselves, the proportions of each land cover within two adjacent mixed pixels as the sub-pixel weight parameters will be inputed, which enhanced the spatial attraction model; The distance function is also a reasonable expression of the non-linear relationship at a distance about the interaction among the pixels. Following an initial random allocation of sub-pixels, the algorithm works in a series of iterations, each of which can optimize the attraction relationship among the sub-pixels, by this the algorithm can improve the spatial dependence among the pixels. This algorithm is tested on SPOT image data at S=5 scale factor, and four land covers are mapped, including water, vegetation, paddy field and urban. The result shows that, this algorithm works reasonably well in multiple classes mapping.