Recently, a number of scientists have found significant evidence that gamma-ray bursts repeat on time scales of days to months (Quashnock and Lamb 1993, 1994; Wang and Lingenfelter 1993, 1995; Strohmayer, Fenimore, and Mirales 1994; Quashnock 1995). Such behavior is similar to the behavior of soft gamma-ray repeaters, as we described earlier.
Repeating sources appear as clusters of bursts on the sky because the
positions of the bursts are known only to about 
(see Figure
19). The evidence for repeating has come from nearest-neighbor
(Quashnock and Lamb 1993, 1994) and angular-temporal correlation (Wang
and Lingenfelter 1993, 1995) analyses of the first BATSE catalog, and a
model comparison study using the same catalog (Strohmayer, Fenimore,
and Mirales 1994). Brainerd et al. (1994) investigated numerous
possible systematic effects and concluded that the results of these
groups is not due to instrumental effects.
Studies of the second BATSE catalog have not confirmed repeating (Brainerd et al 1995; Meegan et al. 1995; Quashnock 1995) but this is expected given the limitations in the second BATSE catalog due to the failure of the tape recorders on board the Compton Observatory (Wang and Lingenfelter 1995; Quashnock 1995; Quashnock and Lamb 1995). A likelihood analysis of the first and second BATSE catalogs, in fact, shows that the odds favoring repeating is as large in the second catalog as in the first, showing that inclusion of the second year of data (which was badly affected by loss of the tape recorders) neither strengthens nor weakens the evidence in favor of repeating. NASA and the BATSE team have worked hard to surmount the difficulties stemming from the loss of the tape recorders, and it is expected that the third BATSE catalog will provide a true test of the repeating hypothesis.
Repeating is naturally expected, even required, in high velocity
neutron star models of gamma-ray bursts, since the total number of
neutron stars in the Galactic corona implies that each must burst
times or more in its lifetime in order to account for the
number of bursts seen per year (Podsiadlowski, Rees, and Ruderman
1995; Lamb, Bulik, and Coppi 1995).
In contrast, cosmological models face severe difficulties in accounting for repeating. The amount of energy needed to power the bursts is so large that in the most widely studied models, such as coalescing neutron stars and failed supernova, it requires the destruction of the source.
Hence, repeating clearly favors the Galactic hypothesis; but we will assign it only modest weight, pending confirmation of it using the data in the third BATSE catalog.