A growing body of research suggests that perception is a probabilistic inference process, in which imprecise sensory inputs are interpreted in light of expectations derived from experience. Optimal (Bayesian) inference enhances perceptual accuracy on average but leads to illusions when stimuli violate expectation. Illusions have been frequently investigated in vision but occur as well in other sensory modalities. Tactile perception is prone to perceptual length contraction illusions, characterized by the dramatic underestimation of the distance between punctate stimuli delivered in rapid succession. Following an introduction to the Bayesian perceptual framework, I will show that a Bayesian observer model with a low-speed expectation replicates perceptual length contraction and related illusions. Time-permitting, I will present the results of psychophysical experiments that my lab has conducted to test the predictions of our Bayesian observer model.