Similar earthquake pairs recorded by the Anza Seismic Network in southern California are used as repeatable sources to place an upper limit on temporal changes in seismic velocity which occurred in the vicinity of the Anza seismic gap in the last 9 yr. Relative arrival times for each pair of events are found using a cross-correlation method and relative locations are calculated to verify that the pairs have nearly identical hypocenters. The time separation between events in these pairs varies from less than a day to almost 7 yr. The long-term changes in seismic travel times, as measured from the pairs with the longest time separation, are not significantly greater than the noise level estimated from the short-time-separation event pairs. Almost all P-wave paths show less than 0.06% (0.007 sec) change in travel time and all S-wave paths have less than 0.03% (0.004 sec) change. Sensitivity tests place an upper bound on travel-time changes that could be compensated by hypocenter mislocation at 0.2%. There is no evidence that localized stress accumulation causes measurable changes in seismic velocity in the Anza region.