This volume assesses approaches to the construction ofcomputer vision systems. It shows that there is a spectrumof approaches with different degrees of maturity androbustness. The useful exploitation of computer vision inindustry and elsewhere and the development of the disciplineitself depend on understanding the way these approachesinfluence one another. The chief topic discussed is autonomy. True autonomy may notbe achievable in machines in the near future, and theworkshop concluded that it may be more desirable - and iscertainly more pragmatic - to leave a person in theprocessing loop. The second conclusion of the workshop concerns the manner inwhich a system is designedfor an application. It was agreedthat designers should first specify the requiredfunctionality, then identify the knowledge appropriate tothat task, and finally choose the appropriate techniques andalgorithms. The third conclusion concerns the methodologies employed indeveloping vision systems: craft, engineering, and scienceare mutually relevant and contribute to one another. The contributors place heavy emphasis on providing thereader with concrete examples of operational systems. Thebook is based on a workshop held as part of the activitiesof an ESPRIT Basic Research Action.