In this paper, sensor fusion system applied to the location of a mobile robot is presented. The idea behind this work is to improve the accuracy in estimating the robot position with respect to systems currently used, which are based on deterministic odometry models. The mainstreaming of sensor fusion involves working with probabilistic mathematical models, which are much better suited to deal with the dynamics of complex environments. A small differential mobile robot with two accelerometers, two odometers and a gyroscope which provide the necessary data to update the estimates provided by the motion model is used. The fusion process is performed using an extended Kalman filter that requires the movement model, the measuring model of the sensors and the set of sensory measurements available in each time instant. The results indicate that the sensor fusion system is more accurate than the reference odometry system. A quantitative analysis shows that in all evaluated cases, the system reports a 38% improvement in estimating the endpoint and 27% in the accuracy over the entire trajectory.