A patch-based super-resolution algorithm is proposed to enlarge a face image from a single low resolution input. Inspired by the property of highly structured faces, the training patches are selected and weighted based on both patch appearance and patch position. The selected patches with certain weights are called bilateral patches, which are incorporated into data consistency to reconstruct the high resolution face image in a patch-wise fashion. Experimental results demonstrate the superiority of the proposed method to some state-of-the-art methods in both visual and quantitative comparisons.