A series of experiments were devised in order to examine aspects of human visual performance during simulated self-motion. The experimental stimuli were computer simulations of observer translational motion through a 3-D random dot cloud. Experiments were specifically designed to obtain data regarding the problem of bias in judgments of heading, and to determine the influence of various experimental factors upon the bias. A secondary aim was to use these results to develop a workable computer model to predict such bias in heading estimation. Heading bias has been known for many years, but it is generally assumed only to be a problem for complex observer motion. However, the current work involved simple observer translation, and found a significant amount of heading bias. A wide variety of experimental factors were examined, and it was found that scene depth and speed had the greatest effect upon the accuracy of heading estimates, with a faster speed or smaller depth reducing bias. It was proposed that yaw eye movements, driven by the rotational component of radial flow, were responsible for the bias. An adaptation of the Perrone (1992) model of heading was used to model this, and a highly significant correlation was obtained between the experimental data and the predictions of the model.