The emerging technology of Radio Frequency IDentification (RFID) has enabled a wide range of automated tracking and monitoring applications. However, the process of interrogating a set of RFID tags usually involves sharing a wireless communication medium by an RFID reader and many tags. Tag collisions result in a significant delay to the interrogation process, and such collisions are hard to overcome because of the limited capabilities of passive RFID tags and their inability to sense the communication medium. While existing anti-collision schemes assume reading all tags at once which results in many collisions, we propose a novel approach in which the interrogation zone of an RFID reader is divided into a number of clusters (annuli), and tags of different clusters are read separately. Therefore, the likelihood of collisions is reduced as a result of reducing the number of tags that share the same channel at the same time. In this paper, we consider two optimization problems whose objective is minimizing the interrogation delay. The first one aims at finding the optimal clustering scheme assuming an ideal setting in which the transmission range of the RFID reader can be tuned with high precision. In the second one, we consider another scenario in which the RFID reader has a finite set of discrete transmission ranges. For each problem, we present a delay mathematical analysis and devise an algorithm to efficiently find the optimal number of clusters. The proposed approach can be integrated with any existing anti-collision scheme to improve its performance and, hence, meet the demand of large scale RFID applications. Simulation results show that our approach makes significant improvements in reducing collisions and delay.