Double-stranded RNA (dsRNA) adenosine deaminase (dsRAD) converts adenosines to inosines within dsRNA. A great deal of evidence suggests that dsRAD or a related enzyme edits mammalian glutamate receptor mRNA in vivo. Here we map the deamination sites that occur in a truncated glutamate receptor-B (gluR-B) mRNA after incubation with pure Xenopus dsRAD. We find remarkable similarities, as well as distinct differences, between the observed deamination sites and the sites reported to be edited within RNAs isolated from mammalian brain. For example, although deamination at the biologically relevant Q/R editing site occurs, it occurs much less frequently than editing at this site in vivo. We hypothesize that the similarities between the deamination and editing patterns exist because the deamination specificity that is intrinsic to dsRAD is involved in selecting editing sites in vivo. We propose that the observed differences are due to the absence of accessory factors that play indirect roles in vivo, such as binding to and occluding certain sites from dsRAD, or promoting the RNA structure required for correct and efficient editing. The work reported here also suggests that dsRAD is capable of much more selectivity than previously thought; a minimal number of deamination sites (average < or = 5) were found in each gluR-B RNA. We speculate that the observed selectivity is due to the various structural elements (mismatches, bulges, loops) that periodically interrupt the base paired region required for editing.