Reliable sensor node localization is a critical and difficult task in a large number of wireless sensor networks applications. Received signal strength indication (RSSI) measurements are a simple and inexpensive way to localize mobile robots, but they suffer from large errors due to noise, occlusions, and multi-path specially in indoor environments. Kalman filters and their variations are widely adopted in the community, although the inherent nonlinearity of the problem suggests the use of more general Bayesian techniques. In this paper, we take on the range-only probabilistic localization problem by a mobile robot that depends on RSSI measurements its only information of a sensor node's location. We propose a method based on a general Probabilistic Graphical Model and our main contribution is the definition of a reasonable, albeit simple, probabilistic RSSI likelihood model connecting distance to the observed RSSI values. Sensor position estimation is performed by integrating a standard robot's position estimator to our Bayesian estimator. Preliminary experimental evaluation indicates that our methodology leads to low error in sensor node's position estimation. Experiments were performed both by simulation, in the Player/Stage platform, and in a real world scenario, using a 802.11g sensor node and a Pioneer 3AT mobile robot.