Acquiring knowledge from examples is frequently used in expert systems. A common model is building of a decision tree which discriminates each class from every other class. Though such a model performs well as far as classification accuracy is concerned, the resultant knowledge is opaque to the user. In this paper, we propose a new model of acquiring knowledge from examples. In this model, a reference class description is first leant from which each class description is learnt. Each of these class descriptions is used to classify test examples. The proposed model has been tested on two applications. The results of these experiments suggest that it is possible to learnt a knowledge base which not only performs well but that is also intelligible.