In a wireless sensor network (WSN), the sensor nodes (SNs) generally localize themselves with the help of anchors that know their own positions. In this setting, the localization process has a high risk of being subverted by malicious anchors that lie about their own position and/or distance from the SNs. In this paper, we propose an efficient scheme that helps the SNs identify these malicious anchors and discard them from the localization process. We introduce the concept of the bound circle of an anchor with respect to an SN as the circle whose center is at the anchor and whose radius is the estimate of the distance between the anchor and the SN. Two bound circles may intersect, resulting in at most two intersection points, of which at least one point is close to the true position of the SN, such a point is defined as a proximal point. Pairwise intersection of bound circles results in a dense cluster of proximal points around the position of the SN. This is true even when some of the anchors used by an SN for localization are malicious and are colluding with an aim to have the SN localized at a false position. We propose CluRoL, a technique that helps each SN to localize itself accurately, using a clustering mechanism that performs clustering of these proximal points. Using the resulting cluster the SN is able to identify the false anchors and exclude them from its localization process. Our technique is decentralized and can be easily used by the standard sensors. Simulation results indicate that when the malicious anchors are not colluding CluRoL can identify on an average more than 72% of them. CluRoL performs even better when the malicious anchors are colluding in an attempt to localize an SN at a false position, identifying more than 85% of the malicious anchors. CluRoL also has very low false positives.