Modern semiconductor manufacturing is very complex and expensive. Maintaining high quality and yield enhancement have been recognized as important factors to build core competences for semiconductor manufacturing companies. Data mining can find potentially useful information from huge databases. This paper proposes a data-mining framework based on decision-tree induction for improving the yield of the solder bumping process in which the various (physical and chemical) input variables that affect the bumping process exhibit highly complex interactions. We conducted an empirical study in a semiconductor fabrication facility in Taiwan to validate this approach. The results show that the proposed approach can effectively derive the causal relationships among controllable input process factors and the target class to enhance the yield. Copyright © 2007 John Wiley & Sons, Ltd.