Spectral lines provide a great deal of information about the emitting region. Missions prior to BATSE---Konus (Mazets et al. 1982), HEAO-1 (Hueter et al. 1987) and Ginga (Murakami et al. 1988; Graziani et al. 1992)---detected absorption lines between 10 and 100 keV which were attributed to cyclotron resonant scattering (which scatters photons out of the line of sight) in 1012 gauss fields (Alexander and Meszaros 1989; Wang et al. 1989). Since neutron stars are the only known anchors for fields of this strength, these observations supported the hypothesis that bursts originate on local neutron stars. However, BATSE has thus far not detected any lines (Palmer et al. 1994; Band et al. 1996) and consequently there has been little interest in explaining spectral lines in a cosmological burst model. A large and stable (temporally and spatially over the absorption region) magnetic field is necessary to produce the narrow lines observed by Ginga, and creating such a field configuration in a relativistic fireball will be a major theoretical challenge.
For the past seven years the BATSE spectroscopy team has been searching for lines and evaluating the results of this search. The reports of the absence of a BATSE detection (Palmer et al. 1994; Band et al. 1996) were based on a visual inspection of spectra. A more comprehensive computerized search has been carried out (Briggs et al. 1996b) and promising line candidates have been identified which are now being evaluated; we expect to issue a definitive report in the next year.
The question of whether the BATSE nondetections are consistent with the detections by previous missions led me to develop a statistical methodology to study the consistency between the results of these missions (Band et al. 1994). For this statistical analysis detailed information is required not only about the detections but also about the nondetections; such data are available from BATSE and Ginga. This methodology can also extract other physical information, such as the likely frequency with which lines occur. The methodology requires simulations of a detector's ability to detect lines (Band et al. 1995) and models for the occurrence of lines within a burst (Band et al. 1997). Thus far only preliminary results have been extracted, in part because only a subset of the necessary Ginga data has been processed (Fenimore et al. 1993b). With various approximations, I find that the two missions are consistent at the few percent level. However, it is also clear that lines are not very common (i.e., they may be present in only a few percent of all bursts).