Electricity theft is a widespread problem that causes tremendous economical losses for all utility companies around the globe. As many countries struggle to update their antique power systems to emerging smart grids, more and more smart meters are deployed throughout the world. Compared with analog meters which can be tampered with by only physical attacks, smart meters can be manipulated by malicious users with both physical and cyber attacks for the purpose of stealing electricity. Thus, electricity theft will become even more serious in a smart grid than in a traditional power system if utility companies do not implement efficient solutions. The goal of this paper is to identify all malicious users in a neighborhood area in a smart grid within the shortest detection time. We propose an Adaptive Binary Splitting Inspection (ABSI) algorithm which adopts a group testing method to locate the malicious users. There are two considered inspection strategies in this paper: a scanning method in which users will be inspected individually, and a binary search method by which a specific number of users will be examined as a whole. During the inspection process of our proposed scheme, the inspection strategy as well as the number of users in the groups to be inspected are adaptively adjusted. Simulation results show that the proposed ABSI algorithm outperforms existing methods.