An interest and development of indoor localization has grown along with the scope of applications. In a large and crowded indoor venue, the population density of access points (APs) is typically much higher than that in small places. This may cause a client device such as a smartphone to capture an imperfect Wifi fingerprints (FPs), which is essential piece of data for indoor localization. This is due to the limited access time allocated per channel and collisions of responses from APs. It results in an extended delay for localization and a massive unnecessary traffic in addition to a high estimation error. This paper proposes a fast and accurate indoor localization method for large-scale indoor venues using a small subset of APs, called representative APs (rAPs). According to our experimental study in a large venue with 1,734 APs, the proposed method achieves the estimation error of 1.8∼2.1m, which can be considered a very competitive performance even in small-scale places with a few hundreds of APs.