ADAR2 is a double-stranded RNA-specific adenosine deaminase involved in the editing of mammalian RNAs by the site-specific conversion of adenosine to inosine. We have demonstrated previously that ADAR2 can modify its own pre-mRNA, leading to the creation of a proximal 3'-splice junction containing a non-canonical adenosine-inosine (A-I) dinucleotide. Alternative splicing to this proximal acceptor shifts the reading frame of the mature mRNA transcript, resulting in the loss of functional ADAR2 expression. Both evolutionary sequence conservation and mutational analysis support the existence of an extended RNA duplex within the ADAR2 pre-mRNA formed by base-pairing interactions between regions approximately 1.3-kilobases apart in intron 4 and exon 5. Characterization of ADAR2 pre-mRNA transcripts isolated from adult rat brain identified 16 editing sites within this duplex region, and sites preferentially modified by ADAR1 and ADAR2 have been defined using both tissue culture and in vitro editing systems. Statistical analysis of nucleotide sequences surrounding edited and non-edited adenosine residues have identified a nucleotide sequence bias correlating with ADAR2 site preference and editing efficiency. Among a mixed population of ADAR substrates, ADAR2 preferentially favors its own transcript, yet mutation of a poor substrate to conform to the defined nucleotide bias increases the ability of that substrate to be modified by ADAR2. These data suggest that both sequence and structural elements are required to define adenosine moieties targeted for specific ADAR2-mediated deamination.