Even at very low template (guest) concentrations, the optimal template-assembled host from a dynamic combinatorial library (DCL) of host fragments may be unobtainable because hetero-oligomers will always be present at higher concentrations than isoenergetic homo-oligomers. Recursively enriched dynamic combinatorial libraries (REDCLs) offer a general solution to this problem that should be applicable to any self-selecting system under thermodynamic control. The utility of the RED CL strategy is demonstrated by determination of the optimal hydrophobic core packing in a template-assembled triple helical protein for which the template is a metal ion and the contributing host fragments are components of a 36-member conformationally restricted peptide library in which each peptide is augmented with a metal-binding moiety. Convergence of the 8436-member DCL to 5 optimal timers (0.06% of the DCL) is complete after four cycles of enrichment. The core packing of the optimal sequences is shown to be native-like, and to reflect the hydrophobic amino acid preferences found in natural parallel three-stranded coiled coils. The influence of potentially critical amino acids on the outcome of the recursive enrichment is explored in a second REDCL. The same peptide sequences were returned and were shown to populate seven of the 8436 possible trimers, or 0.08% of the DCL.