Indoor localization has been the subject of many studies and still remains a very difficult issue due to the presence of multiple paths, fading, shadowing and timely variations in the environment. This paper studies the use of the Received Signal Strength Indicator (RSSI) metric for range estimation in a wireless sensor network. We perform experimental studies which show that, despite a moderate antenna directivity (confirmed by experience in an anechoic chamber), the signal attenuation depends on both of the anchor node and the target node antennas orientation. We also show that the log normal shadow model, often used to model indoor propagation, is not accurate enough to capture irregularities in an office environment. In a 4m×3m office with five anchor nodes and one target node, a confidence interval of about 1m is reached by a maximum likelihood-based localization, only slightly better than random localization.