Accurately localize a mobile vehicle with an easy and quickly deployable system can be very useful for many applications. Herein we present an EKF-SLAM algorithm which allows using radio frequency (RF) beacons without any prior knowledge of their location. As RF beacons provide only range information, recovering their positions is not an easy task. For this range-only SLAM case, a new procedure to instantiate the beacons in the filter is proposed. The method uses two range measurements from different robot's positions to initialize two hypotheses for the beacon's location, which are then integrated in the filter as a Gaussian mixture. This approach is evaluated and compared to other initialization techniques in simulation and with a real dataset. The results show that our approach performs as well as the other existing methods for both trajectory and map errors with a low computational cost.