How we estimate the passage of time is an unsolved mystery in neuroscience. Illusions of subjective time provide an experimental access to this question. Here we show that time compression and expansion of visually marked intervals result from a binding of temporal interval markers. Interval markers whose onset signals were artificially weakened by briefly flashing a whole-field mask were bound in time towards markers with a strong onset signal. We explain temporal compression as the consequence of summing response distributions of weak and strong onset signals. Crucially, temporal binding occurred irrespective of the temporal order of weak and strong onset markers, thus ruling out processing latencies as an explanation for changes in interval duration judgments. If both interval markers were presented together with a mask or the mask was shown in the temporal interval center, no compression occurred. In a sequence of two intervals, masking the middle marker led to time compression for the first and time expansion for the second interval. All these results are consistent with a model view of temporal binding that serves a functional role by reducing uncertainty in the final estimate of interval duration.