This paper is concerned with the detection of “features” of visual patterns, in particular with a view to the construction of a reading automaton. The features can be divided into micro‐features and macro‐features. It is much more difficult to detect a macro‐feature than a micro‐feature of a pattern. For the detection of at least a large class of macro‐features interative processes seem to be the most appropriate, since by these processes the detection of a macro‐feature is reduced to the repeated detection of a micro‐feature, while further a given micro‐feature can be displaced, by such a process, over virtually the whole screen on which the image is projected. Two examples of such processes are discussed.