Indoor positioning has gained considerable attention over the last decade. For time-of arrival based systems, operating in urban and indoor environments the availability of line-of-sight signals is not always guaranteed due to physical obstructions such as buildings, walls, elevator shafts, etc. Specifically non-line-of-sight (NLOS) conditions introduce a bias to the distance estimation which results in significant localization errors. It is therefore vital to be able to identify NLOS channels accurately in real-time and mitigate their impact on location estimation. Channel-based metrics such as RMS delay spread and kurtosis have been proposed for Ultra Wideband systems to identify NLOS channels. Their performance under different system bandwidths (e.g. WLAN), however, has not been investigated in the literature. In this paper we introduce a novel NLOS identification metric based on the entropy of the Channel Impulse Response (CIR) which exhibits distinct statistical properties in different channel conditions. Using frequency domain wireless propagation measurements in a typical indoor environment we show that the entropy metric has a higher detection/identification rate compared to RMS delay spread and kurtosis. We provide empirical evaluation of the test metrics under different system bandwidths and illustrate that entropy shows significant performance gains especially at lower bandwidths. The performance of NLOS identification depends on the accuracy of the estimated metric. Thus in order to estimate the entropy metric accurately we have selected the autoregressive (AR) modeling method which has shown accurate and consistent performance. Finally, we present an investigation of the impact of AR modeling parameters on the performance of entropy-based NLOS identification.