Most transformational linguists would no longer create explicit deep structures. Instead they adopt a surface-interpretive approach. We find deep structures indispensable for projection into a semantic network. In conjunction with a reference architecture based on constraint-posting, they minimize referential non-determinisms. We extend Marcus’ Determinism Hypothesis to include immediate reference, a foundational subc!ass of reference. This Referential Determinism Hypothesis, constitutes a semantic constraint on theories of syntactic analysis, arguing for theories that minimize referential non-determinism. We show that our combination of deep structures and constraint-posting eliminates non-determinism in immediate reference. We conclude that constraint-posting, deep-structure parsers satisfy the referential determinism hypothesis.