This paper presents a novel approach to partial deduction (PD) of logic programming,which is based on abstract interpretations (AIs).In asingle preprocessing phase,given a program P and a goal G, PD and AI for P ∪ {G} are carried out simultaneously,with AI controlling the process of auto-unfolding in PD. PD always terminates if the abstract domain of AI is finite. Moreover, P' has got the information (collected by AI) about its operationalbehavior, which can be used for further optimization of P'.