Radio frequency identification (RFID) technology has been gaining growing interest from a wide range of industries because of its lower cost and increased capability. In this paper, we study the performance of RFID localization in an indoor environment. In particular, we investigate the k-Nearest Neighbour (k-NN) algorithm in different indoor radio propagation environments and different deployment scenarios and settings. By modelling these various indoor features, the k-NN algorithm is tested and observed to be robust in terms of the estimation errors in various indoor localization scenarios. On the other hand, this study reveals that its performance can be improved by tuning certain parameters, e.g. the k value and different weighting methods adaptively for different locations.