A major limitation of current workflow systems is their lack of supporting dynamic workflow modifications. However, this functionality is a major requirement for next-generation systems in order to provide sufficient flexibility to cope with unexpected situations and failures. For example, our experience with data intensive medical domains such as cancer therapy shows that the large number of medical exceptions is hard to manage for domain experts. We therefore have developed a rule- based approach for partially automated management of semantic exceptions during workflow instance execution. When an exception occurs, we automatically determine which running workflow instances w.r.t. which workflow regions are affected, and adjust the control flow. Rules are being used to detect semantic exceptions and to decide which activities have to be dropped or added. For dynamic modification of an affected workflow instance, we provide two algorithms (drcd-and p-algorithm) which locate appropriate deletion or insertion points and carry out the dynamic change of control flow.