After having made a decision in a fine motion discrimination task, subjects' perceived motion direction is systematically biased away from the decision boundary. These repulsive biases were considered the result of a neural read-out mechanism that is optimized for the specific discrimination task, relying on the most informative neurons whose tuning preferences are slightly away from the decision boundary (Jazayeri and Movshon, 2007). Based on this explanation we would expect the biases to disappear if subjects were given the correct answer to the discrimination task instead of performing the task themselves. We conducted two psychophysical experiments to test this prediction. Experiment 1 was aimed at replicating the effects of the original study for an orientation discrimination task. We used stimuli that consisted of an array (N=24) of small line segments (length of 0.7 degs) shown within a circular aperture (diameter of 5 degs). The orientation of each line segment was drawn from a Gaussian centered on the true stimulus orientation. Stimulus orientations were uniformly sampled over a range +/- 21 degrees relative to a randomly chosen reference orientation. Subjects first had to indicate whether the stimulus orientation was clockwise (CW) or counter-clockwise (CCW) of the reference orientation. Subsequently they had to reproduce the stimulus orientation. Experiment 2 was identical to Experiment 1 except that subjects were given the correct stimulus' category (CW/CCW) at the beginning of each trial. We found nearly identical repulsive biases in both experiments. Hence, we conclude that the repulsive biases cannot be the result of a task-dependent neural read-out mechanism as previously suggested. Rather, we argue that they occur in both experiments because humans i) treat their own perceptual decisions as if they were true and ii) condition their subsequent estimates accordingly as predicted by a ``self-consistent'' Bayesian observer model (Stocker and Simoncelli, 2008).