Abstract: A partial least squares (PLS)-based super-resolution method is proposed for fast high resolution facial image reconstruction. First, projects all the high-resolution and low-resolution images is projected onto their respective eigen-space, and then the PLS regression model is built by capturing the statistical relationship between those projection coefficient pairs. When the low resolution input is given, the corresponding high-resolution image can be derived from the well trained PLS regression model easily. Experiments showed that the proposed method can achieve satisfying result with high speed if only the off-line training is taken.