The photointerpretation process is one of the most difficult, but seemingly feasible, processes to automate. The non-numeric nature of the basic input medium, the great variation in imagery representing a single target class, the absence of satisfactory mathematical models for background noise (which are often patterns of poteutial, but not immediate interest) as well as target class patterns, and the intuitive logical requirement of higher order inferential decisions, make this problem one of the most interesting potential applications for computers today. This paper characterizes the photointerpretation process in terms of the input pattern structure and an overall decision tree for the process. A general model for pattern recognition is presented and related to the photointerpretation process. The use of the general-purpose computer as a research tool to permit study of automatic photointerpretation is illustrated by several specific examples. Nonlinear, two-dimensional filters which were implemented and evaluated entirely within an IBM-7044 are examples treated, while experiments on linear discrimination pattern recognizers for use in photointerpretation represent an opposite extreme. The results of an extensive experimental program to explore linear discriminator pattern recognizers show that automatic photointerpretation is more heavily dependent on preprocessors than on image classifiers. These results are discussed in terms of 1) their significance to ultimate solution of current problems, 2) the implication of the accomplishments to date, and 3) recommended directions for research in this area.