Vibrotactile tilt feedback was used to help vestibulopathic subjects control their anterioposterior (AP) sway during sensory organization tests 5 and 6 of Equitest computerized dynamic posturography. We used four kinds of signals to activate the feedback. The first signal was proportional (P) to the measured tilt of the subject, while the second used the first derivative (D) of the tilt. The third signal was the sum of the proportional and one half of the first derivative signals (PD). The final signal used a prediction of the subject's sway projected 100 msec in advance. The signals were used to activate vibrators mounted on the front of the torso to signal forward motion, and on the back of the torso for backward motion. Subject responses varied significantly with the kind of feedback signal. Proportional and derivative feedback resulted in similar root mean squared tilt, but the PD signal significantly reduced the tilt compared to either P or D feedback. The predicted motion signal also reduced the response compared to the PD signal. These preliminary results are somewhat consistent with an inverted pendulum model of postural control, but need to be confirmed with a larger study that also considers mediolateral tilt and feedback. The improvement by using a predictor is consistent with compensating for a neural processing delay.