In ancient Greek, relative pronouns are, as a rule, subject to wh-movement and obligatorily surface at the left edge of the relative clause. However, the archaic poet Pindar sometimes allows material belonging to the relative clause to appear in front of the relative pronoun, which is then postponed within its clause. In this paper, I survey all relative clauses in the surviving texts by Pindar and study the possible differences in semantics and syntax between relative clauses with initial and postponed relative pronouns, which turn out to be indistinguishable in both respects. I suggest that postponed relative pronouns do move syntactically to the Spec of their relative clause but are then optionally treated as postpositive words and surface in second position in the relative clause. Phonological arguments, based on the distributional properties of postpositive words and on the metrical makeup of Pindar’s texts, are put forward to show how postponed relative pronouns select a host at the left edge of the relative clause and incorporate phonologically to it. The informational status of relative pronouns as ratified (given) topics triggers their phonological demoting, which turns them into postpositive words, a regular process in ancient Greek. Approaching the position of relative pronouns as a conflict between syntactic and (informationally driven) phonological alignment explains why Pindar’s strategy for relativization remained rare in ancient Greek and eventually disappeared: It took one specific poetic genre to allow phonology to outrank syntax.