Although predation by fishes is thought to structure benthic invertebrate communities on coral reefs, evidence to support this claim has been difficult to obtain. We deployed an array of eight sponge species on Conch Reef (16 m depth) off Key Largo, Florida, USA, and used a remote video-camera to record fish activity near the array continuously during five daylight periods (6 h for 1 d, at least 11.5 h for 4 d) and one night period (11 h). Of the eight sponge species, four were from adjacent reefs (Agelas wiedenmayeri, Geodia neptuni, Aplysina fistularis, and Pseudaxinella lunaecharta), and four were from a nearby mangrove habitat (Chondrosia collectrix, Geodia gibberosa, Halichondria sp., andTedania ignis). Each species of reef sponge was chosen to match the corresponding mangrove species in form and color (black, brown, yellow, and red, respectively). Predation events only occurred during daylight hours. Tallies of the number of times fishes bit sponges revealed intense feeding by the expected species of sponge-eating fishes, such as the angelfishHolacanthus bermudensis, H. tricolor, andPomacanthus arcuatus, the cowfishLactophrys quadricornis, and the filefishCantherhines pullus, but surprisingly also by the parrotfishSparisoma aurofrenatum andS. chrysopterum. Of 35 301 bites recorded, 50.8% were taken by angelfish, 34.8% by parrotfish, and 13.7% by trunkfish and filefish. Mangrove sponges were preferred by all reef fishes; 96% of bites were taken from mangrove species, with angelfish preferringChondrosia collectrix and parrotfish preferringGeodia gibberosa. Fishes often bit the same sponge repetitively, and frequently consumed entire samples within 30 min of their deployment. Sponge color did not influence fish feeding. Two of the four mangrove sponge-species deployed on the array were also found living in cryptic habitats on adjacent reefs and were rapidly consumed by fishes when exposed. Our results demonstrate the importance of fish predation in controlling the distribution of sponges on Caribbean reefs.