Influenza pandemics occur unpredictably when zoonotic influenza viruses with novel antigenicity acquire the ability to transmit amongst humans1. Host range breaches are limited by incompatibilities between avian virus components and the human host. Barriers include receptor preference, virion stability and poor activity of the avian virus RNA-dependent RNA polymerase in human cells2. Mutants of the heterotrimeric viral polymerase components, particularly PB2 protein, are selected during mammalian adaptation, but their mode of action is unknown3, 4, 5, 6. We show that a species-specific difference in host protein ANP32A accounts for the suboptimal function of avian virus polymerase in mammalian cells. Avian ANP32A possesses an additional 33 amino acids between the leucine-rich repeats and carboxy-terminal low-complexity acidic region domains. In mammalian cells, avian ANP32A rescued the suboptimal function of avian virus polymerase to levels similar to mammalian-adapted polymerase. Deletion of the avian-specific sequence from chicken ANP32A abrogated this activity, whereas its insertion into human ANP32A, or closely related ANP32B, supported avian virus polymerase function. Substitutions, such as PB2(E627K), were rapidly selected upon infection of humans with avian H5N1 or H7N9 influenza viruses, adapting the viral polymerase for the shorter mammalian ANP32A. Thus ANP32A represents an essential host partner co-opted to support influenza virus replication and is a candidate host target for novel antivirals.