This article considers the question of a priori assessment of the safety of releasing recombinant DNA engineered organisms. Now and for the foreseeable future, decisions to release such an organism must be based on the results of limited, case-by-case risk assessment studies. The criteria calling for the termination of release programs must be agreed upon in advance of these studies. There is no justification for excluding classes of release organisms from risk assessment. Theory is useful in suggesting a hierarchy of risks, raising the questions that have to be addressed in case-by-case risk assessment and providing protocols for the standardization and execution of these studies. We do not believe that theory can be used to argue categorically for or against the safety of specific releases of recombinant DNA engineered organisms.