Subjects performed a three-dot alignment in the frontoparallel plane. We found systematic deviations in alignment, especially for diagonally oriented stimuli. The biases did not depend on the angular size of the stimuli which was varied between 0.8 and 20 deg. We put forward a tentative explanation based on saccade trajectories. Extending the task to judgements of the straightness of virtual lines consisting of a varying number of dots showed that the biases decrease gradually when the number of dots increases. This suggests that there are two different and competing mechanisms to judge the straightness of virtual lines.