In this paper a haptic matching task is used to analyze haptic spatial processing. In various conditions, blindfolded participants were asked to make a test bar parallel to a reference bar. This always resulted in large but systematic deviations. It will be shown that the results can be described with a model in which an egocentric reference frame biases the participants' settings: What a participant haptically perceives as parallel is a weighted average of parallel in allocentric space and parallel in egocentric space. The basis of the egocentric reference frame is uncertain. There is strong evidence that at least a hand-centred reference frame is involved, but possibly a body-centred reference frame also plays a role.