Loop–loop interactions among nucleic acids constitute an important form of molecular recognition in a variety of biological systems. In HIV-1, genomic dimerization involves an intermolecular RNA loop–loop interaction at the dimerization initiation site (DIS), a hairpin located in the 5′ noncoding region that contains an autocomplementary sequence in the loop. Only two major DIS loop sequence variants are observed among natural viral isolates. To investigate sequence and structural constraints on genomic RNA dimerization as well as loop–loop interactions in general, we randomized several or all of the nucleotides in the DIS loop and selected in vitro for dimerization-competent sequences. Surprisingly, increasing interloop complementarity above a threshold of 6 bp did not enhance dimerization, although the combinations of nucleotides forming the theoretically most stable hexanucleotide duplexes were selected. Noncanonical interactions contributed significantly to the stability and/or specificity of the dimeric complexes as demonstrated by the overwhelming bias for noncanonical base pairs closing the loop and covariations between flanking and central loop nucleotides. Degeneration of the entire loop yielded a complex population of dimerization-competent sequences whose consensus sequence resembles that of wild-type HIV-1. We conclude from these findings that the DIS has evolved to satisfy simultaneous constraints for optimal dimerization affinity and the capacity for homodimerization. Furthermore, the most constrained features of the DIS identified by our experiments could be the basis for the rational design of DIS-targeted antiviral compounds.