We describe how to diagnose the origin of failure in chemical plants utilizing their qualitative models. We sometimes obtain, however, unsatisfactory diagnostic results when we use the most primitive model called “signed digraph”, which only represents the cause-effect relationships among the state variables in the objective chemical plant. Two improvements are presented in order to give more satisfactory results. One is the improvement by utilizing delays of immediate influences among state variables and the other is by using dynamic information obtained from sensors. We demonstrate effects of the improvements via simulation and experiment.