We hypothesize that resolved agreement with gender-mismatched human and inanimate coordinated nominals is predictable from how animacy is encoded within a language’s gender system, in ways that are restricted by a feature-geometric account à la Harley and Ritter 2002. We demonstrate the success of this account for three languages of different IE branches: Greek, Icelandic, and BCS (Bosnian/Croatian/Serbian). We also explore the implications of our analysis for other languages.