Novel techniques are described for hand-printed numeral recognition and generation of various-sized high-quality graphic characters, which permit efficient communications between facsimile terminals and computer centers. To achieve extremely precise numeral recognition despite the poor characteristics of the images obtained by facsimile terminals, a supplementary preprocessing method and a similar-shaped-character distribution method have been developed. Using these procedures, about 80% of the characters which cannot be recognized by the basic recognition procedure are rendered recognizable. Transmission bit errors affect encoded facsimile signals of an entry sheet in the form of recognition degradation or failure of row detection mark length. The occurrence probability of these kinds of errors is estimated, and it is shown from the degree of recognition degradation that the impact is negligibly small from the standpoint of practical use. A novel smoothing method for code-to-pattern conversion has also been developed to improve the picture quality of graphic character patterns.>