Rocaglamide A (RocA) typifies a class of protein synthesis inhibitors that selectively kill aneuploid tumour cells and repress translation of specific messenger RNAs1,2,3,4. RocA targets eukaryotic initiation factor 4A (eIF4A), an ATP-dependent DEAD-box RNA helicase; its messenger RNA selectivity is proposed to reflect highly structured 5′ untranslated regions that depend strongly on eIF4A-mediated unwinding5. However, rocaglate treatment may not phenocopy the loss of eIF4A activity, as these drugs actually increase the affinity between eIF4A and RNA1,2,6. Here we show that secondary structure in 5′ untranslated regions is only a minor determinant for RocA selectivity and that RocA does not repress translation by reducing eIF4A availability. Rather,in vitroand in cells, RocA specifically clamps eIF4A onto polypurine sequences in an ATP-independent manner. This artificially clamped eIF4A blocks 43S scanning, leading to premature, upstream translation initiation and reducing protein expression from transcripts bearing the RocA–eIF4A target sequence. In elucidating the mechanism of selective translation repression by this lead anti-cancer compound, we provide an example of a drug stabilizing sequence-selective RNA–protein interactions.