We present a novel algorithm for self-localization in sensor networks without any prior knowledge on the locations of the sensors. We assume that all sensors in the network can receive and transmit, thus we obtain time difference of arrival measurements for all combinations of sensors. Using the full set of these differences in arrival times in the network we are able to obtain the relative location of the sensors nodes, the shape of the network. This leaves us with the problem of anchoring the network to its absolute location, which we solve using additional transmitting beacons at known locations. Experimental results from numerical simulation demonstrate the performance of our approach under various conditions.