It is impossible to collect more than a tiny proportion of all of the possible examples of a given hue to form a training set for a machine that learns to discriminate colours. In view of this, it is argued that colour generelization is essential. Three mechanisms for learning colours, as defined by a human being, are described. One of these is based upon an idea developed by A.P. Plummer and is implemented in a commercial device known as the “intelligent camera”. This implementation can learn the characteristics of coloured scenes presented to it and can segment a video image in real-time. This paper presents four procedures that allow the range of colours learned by such a system to be broadened so that recognition is made more reliable and less prone to generating noisy images that are difficult to analyse. Three of the procedures can be used to improve colour discrimination, while a fourth procedure is used when a single and general colour concept has to be learned. Several experiments were devised to demonstrate the effectiveness of colour generelization. These have shown that it is indeed possible to achieve reliable colour discrimination / recognition for such tasks as inspecting packaging and fruit. A practical system based upon the intelligent camera and controlled by software written in PROLOG has been developed by the authors and is being used in a study of methods for declarative programming of machine vision systems for industrial applications.