Retroviruses synthesize a terminally redundant genomic RNA that contains canonical polyadenylation signals at both ends. Production of this RNA requires that the 5' copy of these signals be ignored, while the 3' copy must be utilized. Two models have been presented for how this occurs in the human immunodeficiency virus, HIV: (i) the core HIV poly(A) signals (AAUAAA and a downstream GU-rich element) might be inefficient and require supplementation by activating sequences found only at the 3' end of the RNA; or (ii) cap site proximity might actively suppress polyadenylation at the 5' site. We have examined both possibilities in HIV-infected cells and in cells transfected with a variety of model constructs. We find that infected cells harbor few or no detectable products of 5' polyadenylation; however, the core HIV processing signals can mediate processing fairly efficiently (65%) when positioned at the 3' end of heterologous transcripts. While this efficiency can be further increased to greater than 95% by inclusion of upstream sequences from the viral U3 region, the absence of these U3 signals is insufficient by itself to account for 5' signal bypass. By contrast, the efficiency of these core elements is greatly suppressed when they are positioned within approximately 450 nucleotides of the cap site. This distance-related suppression can be modestly diminished by insertion of U3 sequences between the cap site and HIV poly(A) signal. We suggest that the primary determinant of 5' poly(A) site bypass is cap site proximity; the absence of U3 sequences at this position contributes secondarily to the bypass by enhancing the sensitivity of the pA signal to the suppressive effects of promoter proximity.