We examine the role and potential application areas of machine vision inspection systems in the printed circuit board (PCB) manufacturing industry. Available automatic optical inspection (AOI) techniques are reviewed, this revealing limitations which stem mainly from the stand-alone operational nature of the present-generation AOI systems. As a result the authors argue that, fuelled by the need to achieve integration and information sharing between AOI systems and other components of computerized manufacturing systems, it is necessary to define architectural guidelines that structure the way in which next generation AOI systems should be organized in order to enable them to share information of a common interest and cooperate on an enterprise-wide basis. We also examine typical applications of AOI in PCB manufacture. Upon extracting common features of these applications, a model of AOI task decomposition is proposed, which characterizes the main features of the two major subtasks that an AOI system needs to perform; so that it can benefit from information sharing and improve its performance in terms of providing more valuable and reusable inspection information to other computerized systems and processes. These two subtasks are (1) an initial local inspection subtask, and (2) a global inspection, analysis and inspection information generation subtask. The importance of the second subtask relates to the general requirement for AOI systems not only simply to automate PCB inspection processes, but also to provide readily reusable inspection information that can be utilized to support other relevant manufacturing processes. Based on the AOI task decomposition model, we propose a hierarchical architecture which can structure the design, implementation and re-organisation of AOI software algorithms and routines, in such a way that typical AOI operations can be enhanced, this through making use of available global information, and that more widespread, efficient and timely use of AOI-generated inspection information can be made by other manufacturing processes. Finally; a proof-of-concept implementation of the AOI reference architecture is described.