Protein-primed replication of hepatitis B viruses (HBVs) is initiated by the chaperone dependent binding of the reverse transcriptase (P protein) to the bulged epsilon stem-loop on the pregenomic RNA, and the epsilon-templated synthesis of the 5' terminal nucleotides of the first DNA strand. How P protein recognizes the initiation site is poorly understood. In mammalian HBVs and in duck HBV (DHBV) the entire stem-loop is extensively base paired; in other avian HBVs the upper stem regions have a low base pairing potential. Initiation can be reconstituted with in vitro translated DHBV, but not HBV, P protein and DHBV epsilon (Depsilon) RNA. Employing the SELEX method on a constrained library of Depsilon upper stem variants, we obtained a series of well-binding aptamers. Most contained C-rich consensus motifs with very low base pairing potential; some supported initiation, others did not. Consensus-based secondary mutants allowed to pin down this functional difference to the residues flanking the conserved loop, and an unpaired U. In vitro active consensus sequences also supported virus replication. Hence, most of the upper stem acts as a spacer, which, if not base paired, warrants accessibility of relevant anchor residues. This suggests that the base paired Depsilon represents an exceptional rather than a prototypic avian HBV epsilon signal, and it offers an explanation as to why attempts to in vitro reconstitute initiation with human HBV have thus far failed.