In this paper, we present an automatic method for character prototyping and font type characterization in machine-printed document images at a character level. To do so, we use a generic textural approach, which considers text as a texture, instead of working at a pixel level like most of the methods proposed so far. In this way, Gabor filtering seems to be an appropriate tool for texture characterization, since its design has been inspired by the human visual system. The objective of the paper is then to verify this hypothesis by applying our method on a corpus composed of what we call "typographically rich and recurrent" machine-printed document images.