RFID is an automatic identification technique that stores and remotely retrieves data on small devices called tags. This technology enables tracking of people and objects, supporting the conception of pervasive networks if identities are linked in real-time to their locations. We propose L-VIRT, a 3-D localization method for RFID tags based only on connectivity information. Virtual landmarks are combined with defined topological constraints to localize tags. A relaxation of the constraints is introduced to attenuate the effect of reading errors and, thus, to increase the precision of the localization. Simulation results show a fine-grained accuracy in free space or moderately obstructive environments, and a satisfactory accuracy in very cluttered environments. Moreover, the use of additional mobile readers enables rapid and large localization error reduction in sparse deployments of static readers.