In this paper we present the hardware design and characteristics of a non-conventional text reading system. By non-conventional, we define a text reading system that is able to scan, detect, extract, and recognize the text characters directly from the `page' without sending strings of 0s and 1s to the host computer (which uses software techniques to recognize the text characters). The non-conventional system is also called `character-driven' OCR system. The system proposed here consists of six main parts, a raster scanner unit, the character pre- processing circuit, the FCC generation unit, the line-generator recognizer unit, the graph generator unit and the matching processor unit. In particular, the scanner array (GRAMMA) consists of (12 X 9) photo-cells and is driven by the 3-D motion mechanism to focus and extract the typed text. The gramma array produces a bit-map for the character which is then fed to the pre-processing circuit which applies curve fitting and thinning techniques to produce a normalized skeleton image of the character, which is fed to the FCC generator. The FCC generator produces a generic code string of the character which is passed on to the line generator recognizer unit which breaks the FCC string into straight-line and curve segments. The graph generator produces interrelations between the different segments. The graph form of the character is then fed to the matching processor which produces an appropriate ASCII code of the character. The non-conventional system can recognize typed as well as handwritten characters of different fonts and sizes.