Radio frequency identification (RFID) systems have been widely deployed for various applications such as object tracking, 3-D positioning, supply chain management, inventory control, and access control. This paper concerns the fundamental problem of estimating RFID tag population size, which is needed in many applications such as tag identification, warehouse monitoring, and privacy-sensitive RFID systems. In this paper, we propose a new scheme for estimating tag population size called Average Run-based Tag estimation (ART). The technique is based on the average run length of ones in the bit string received using the standardized framed slotted Aloha protocol. ART is significantly faster than prior schemes. For example, given a required confidence interval of 0.1% and a required reliability of 99.9%, ART is consistently 7 times faster than the fastest existing schemes (UPE and EZB) for any tag population size. Furthermore, ART's estimation time is provably independent of the tag population sizes. ART works with multiple readers with overlapping regions and can estimate sizes of arbitrarily large tag populations. ART is easy to deploy because it neither requires modification to tags nor to the communication protocol between tags and readers. ART only needs to be implemented on readers as a software module.