olving Problems through serially removing differences is a kind of problem decomposition. The order of differences to be removed can be planned based on the analysis of operators. The main idea of this plan is to attain next subgoal without undoing the last attained subgoal. Thus, during problem decomposition, those subgoals which keep the operators most usable are chosen, some what like the least constraint propagation.When a set of hierarchic subgoals is generated, the search of initial problem is to attain a serial of subgoals in the set, during which backtracking may happen.