Early Warning Systems (EWS) are non-structural measures for landslides disaster prevention. They are based on the detection of impending failure signals. The results of a landslide simulation experiment where accelerometers were used to identify pre-failure signals are presented in this paper. Landslide was simulated in a tilting flume filled with sandy soil. During the experiment, the flume was fixed at 30° inclination and water percolated through the soil until it slid. Accelerometers were embedded into the soil and recorded acceleration data from the beginning of the experiment until failure. Acceleration data were analyzed in time domain aiming at estimating translational velocity of the movement. Angular variation was also estimated from acceleration data. The experiment was recorded with a camera and pictures were used for Particle Image Velocimetry (PIV) analysis, in order to validate the estimated translational velocity. Results showed that accelerometers can identify pre-failure signals before any macroscopic movement could indicate impending failure in fast to very fast landslides, showing their potential to be used in EWS. Validation of estimated velocities was not always possible due to PIV setup constraints and the velocity of the mass movement simulated. In fact, the estimated translational velocities seem to be unreliable. On the other hand, the results suggest that acceleration data and angular position variation trend and rate can be incorporated into EWS.