When generating new design concepts, most industrial designers tend to draw upon stereotypical images and their own personal design experiences. The evaluation of each individual design candidate in terms of its ability to meet the demands of the marketplace is a crucial step within the conceptual design stage. In this study, a feature-based hierarchical CAD model is initially constructed, in which the related form parameters are thoroughly defined in applicable domains to facilitate the automatic generation of new product forms. A fuzzy neural network algorithm is then applied to establish the relationships between the input form parameters and a series of adjectival image words. In a reverse process, genetic algorithms are employed to search for a near-optimal design which satisfies the designer's required product image by using the trained neural network as a fitness function. The proposed method represents an automatic design system, which provides designers with the ability to rapidly obtain a product form and its corresponding image, or to search for the ideal form which fits a required image in a shorter lead-time.