The invention discloses a device time mark fault-tolerance processing method in a grid fault diagnosis. The device time mark fault-tolerance processing method comprises the following steps of: starting analysis on a master station fault diagnosis module through a fault diagnosis starting criterion, and collecting protective action events and on-off action events in a fault message acquisition time segment; preliminarily filtrating motion events in the acquisition time segment, dividing the filtrated protective action events and on-off action events in the fault time segment into clusters, and dividing the relative protective action events and on-off action events in into one group according to an association relationship between primary equipment and secondary equipment in a model and real-time network topology, i.e. dividing the clusters and the groups of the motion events in the fault time segment at first; and then judging and selecting a uniform time reference corrected value of each group, using the corrected value as a reference for all protecting devices in each group, and translating a time shaft left and right; and correcting all on-off motion events based on the corrected main protective time of a corresponding master station end, i.e. correcting the time of each group in the cluster.