The first step in failure recovery analysis identifies potential problems in the planner's interaction with its environment; the remaining three steps explain how those problems may have been produced by the planner's actions and suggest redesigns to avoid the resulting failures. The statistical dependencies are mapped to structural dependencies in the planning knowledge bases suspected to be vulnerable to failure. Then, the interactions and vulnerable plan structures are used to generate explanations of how the observed failures might occur. Finally, the explanations serve to recommend redesigns of the planner and recovery component. These steps do not rely on statistical techniques or arguments, so we will not describe them in detail here. Interested readers should consult [Howe 92].