Due to the lack of support on the high resolution image in short time, the land-cover change detection is always applied on the multi-temporal remote sensed images with coarse resolution. They contain a large number of mixed pixels, which can seriously limit the utility of change detection. Pixel unmixing (PU) can be applied to improve this situation through deriving the abundances and generating the fractional change map. Thus it can be regarded as a more reasonable tool to monitor the sub-pixel changes. In this paper a new sub-pixel land-cover change detection method based on PU and EM (Expectation Maximization) algorithm is proposed. The results of experiment on the multitemporal thematic mapper (TM) images show that the proposed method not only can obtain higher change detection accuracy, but also can maintain better geometric properties comparing with the traditional methods.