Current and future ground- and space-based telescopes are capable of producing immense volumes of data which, to make the telescope truly productive, must be analyzed in reasonable periods of time after acquisition. Classical methods of computerized data analysis tend to become more specific as the data volume increases, thereby not fully utilizing the information content of the data. To reverse this trend, we investigate artificial intelligence-based analysis systems to act either as the principal analysis tool for a data set, or to act as a co-processor to a classical statistical analysis system. We have designed, and are in the process of implementing, an image processing system based on concepts of artificial intelligence. The input images are produced by the CCD Transit Instrument (CTI)1. Standard astronomical classification has generally been accomplished by comparison to a hierarchy of standard objects (e.g. the MK spectral classification system2) and, in our new system, we mimic the use of such standards by a network of prototypes. The prototypes are represented within the computer as frames3, each of which contains knowledge either of a standard object or of the links between such objects. Thus the frames provide both the goal of our classification search, and the information required to guide that search. Such an approach provides several major advantages when compared to the classical, statistically-based pattern recognition systems. Firstly, it classifies as an astronomer would, thus giving credibility to its conclusions. Secondly, it provides a natural avenue for machine discovery of new classes of objects thereby meeting one of the major goals of CTI - serendipitous discovery. Thirdly, once set up, it will not make enormous demands upon the user's time.