DEAD-box proteins are the largest family of nucleic acid helicases and are crucial to RNA metabolism throughout all domains of life 1, 2 . They contain a conserved ‘helicase core’ of two RecA-like domains (domains 1 and 2; D1 and D2, respectively), which uses ATP to catalyze the unwinding of short RNA duplexes by nonprocessive, local strand separation 3 . This mode of action differs from that of translocating helicases and allows DEAD-box proteins to remodel large RNAs and RNA-protein complexes without globally disrupting RNA structure 4 . However, the structural basis for this distinctive mode of RNA-unwinding remains unclear. Here, structural, biochemical, and genetic analyses of the yeast DEAD-box protein Mss116p indicate that the helicase core domains have modular functions that enable a novel mechanism for RNA duplex recognition and unwinding. By investigating D1 and D2 individually and together, we find that D1 acts as an ATP-binding domain and D2 functions as an RNA-duplex recognition domain. D2 contains a nucleic acid-binding pocket that is formed by conserved DEAD-box protein sequence motifs and accommodates A-form but not B-form duplexes, providing a basis for RNA substrate specificity. Upon a conformational change in which the two core domains join to form a ‘closed-state’ with an ATPase active site, conserved motifs in D1 promote the unwinding of duplex substrates bound to D2 by excluding one RNA strand and bending the other. Our results provide a comprehensive structural model for how DEAD-box proteins recognize and unwind RNA duplexes. This model explains key features of DEAD-box protein function and affords new perspective on how the evolutionarily related cores of other RNA and DNA helicases diverged to use different mechanisms.