Abstract: A method and system for automatic optical character recognition wherein a continuous flow of image data is processed at a scan rate high compared to the relative movement of the information-containing field across a sensor array. A two-dimensional array of photosensors is self-scanned to provide a continuous flow of binary character segment and void zone representations which are decoded by reference to a ROM to produce multi-bit feature codes characterizing unique relationships between line segments. For each row or column scanned, a representative feature code is produced with a sequence of such codes representing a complete scan of the sensor array. Successive sequences of feature codes are applied to a decision network having a multiplicity of nodes where each node is accessible to every other node by one state change to produce a representative output character code.