Clock accuracy is a basic parameter of any seismic station and has become increasingly important for seismology as the community seeks to refine structures and dynamic processes of the Earth. In this study, we measure the arrival time differences of moderate repeating earthquakes with magnitude 5.0–5.9 in the time range of 1991–2017 at the same seismic stations by cross‐correlating their highly similar waveforms and thereby identify potential timing errors from the outliers of the measurements. The method has very high precision of about 10 ms and shows great potential to be used for routine inspection of the timing accuracy of historical and future digital seismic data. Here, we report 5131 probable cases of timing errors from 451 global and regional stations available from the Incorporated Research Institutions for Seismology Data Management Center, ranging from several tens of milliseconds to over 10 s. Clock accuracy seems to be a prevailing problem in permanent stations with long‐running histories. Although most of the timing errors have already been tagged with low timing quality, there are quite a few exceptions, which call for greater attention from network operators and the seismological community. In addition, seismic studies, especially those on temporal changes of the Earth’s media from absolute arrival times, should be careful to avoid misinterpreting timing errors as temporal changes, which is indeed a problem in some previous studies of the Earth’s inner core boundary.