Managing human–wildlife conflict is often complicated by apparent mismatches between community perceptions and measures of directly incurred losses. Fishers in Agatti Island (Lakshadweep, India) associate recent increases in green turtle (Chelonia mydas) populations with declining fish catches, resulting in targeted killing of turtles. We compared fisher perceptions in Agatti with a very similar atoll, Kadmat, with much lower turtle densities. Nearly 90% of Agatti fishers interviewed blamed turtles for declining catch compared with 20% in Kadmat and proposed two mechanisms for this decline: direct interference (e.g., turtles damaged gear) which we define as first order conflict, and indirect mechanisms (second order conflict): turtles overgrazed seagrasses, thereby reducing fish catch. We evaluated the magnitude of gear loss with interviews and tested proposed indirect mechanisms with a turtle density gradient, before–after comparisons (taking advantage of an increase in turtles in Kadmat and concurrent decrease in Agatti) and a natural herbivore exclosure. These complementary approaches supported fisher-pro- posed second-order mechanisms: at high densities, turtles heavily grazed seagrasses, significantly reduced canopy heights, lowered fish recruit abundance, food fish biomass and catch. Estimates of losses incurred in Agatti show that first-order conflict cost fishers USD 0.6 fisher-1 year􏰄1, while second-order pathways accounted for USD 887 fisher-1 year-1. Our results show that local perceptions are fueled by often-complex mechanisms that, though not always straightforward to measure, are very important in generating conflict. Reconciling the human–wildlife interface requires an adequate accounting of direct and indirect mechanisms to more completely reflect true losses communities bear for living with wildlife.