Medical problems can be hierarchically decomposed into sub problems. In general, the problems are solved by humans. For some very seldom occurring sub problems, e. g., developing a patient management plan for the acute radiation syndrome, or for routine problems, e. g., report generation in laboratory medicine for 95% of the cases, knowledge-bases could perform the task more efficiently than humans. An architecture for medical knowledge-based assistance systems is presented allowing for the co-operation of physicians and knowledge-bases to efficiently solve medical problems. The collaboration is obtained by applying a cognitive problem-solving model implemented on a blackboard control architecture. The architecture has been prototypically implemented and partially evaluated for the medical management of the acute radiation syndrome and in laboratory medicine.