In this paper, following a review of the historical development of research methods for predicting volume sales and brand shares of new products, a new model (MicroTest) is described which uses information gathered in a concept/product test for volume prediction. The model makes use of brand related parameters (such as advertising and distribution), altitudinal predispositions (e.g. experimentalism), and circumstantial factors as input to the model, and these are described, together with the method of integrating these for predicting at the individual respondent level. Individual results are then accumulated across a sample of individuals, and grossed up to provide national sales estimates. The paper describes the various development stages undergone in the construction of the model, and the techniques used to assist this process. In particular, the way in which Artificial Intelligence techniques such as rule induction was used is discussed. Finally, the paper discusses the way in which the basic model may be extended, and some recent work which used the model to generate a measure of cumulative penetration.