Trophic cascades are predicted to occur when the abundance of predators is increased, directly reducing the abundance of the intermediate prey and indirectly increasing the abundance of the prey at the base of a food web. Mixed trophic impact analysis of a network model developed for Apalachee Bay, near St. Marks, FL, USA predicted such a trophic cascade, in that increased abundance of juvenile gulf flounder Paralichthys albigutta ( x¯ = 149 mm SL, effective trophic level 3.9) should have a negative impact on juvenile spot Leiostomus xanthurus ( x¯ = 30 mm SL, effective trophic level 2.9) and a positive impact on benthic polychaetes (effective trophic levels 2.3 for deposit feeders and 3.0 for predatory polychaetes) in Halodule wrightii seagrass beds. We tested the predictions of the mixed trophic impact analysis by manipulating the abundance of the high trophic-level species (juvenile gulf flounder) in a cage-exclusion study in the North River, near Harkers Island, NC, USA. We compared the polychaete communities in St. Marks, FL and Harkers Island, NC, and showed that they are 51% similar (Jaccard's Index) at the family level, with the same eight dominant families (Nereidae, Capitellidae, Syllidae, Spionidae, Cirratulidae, Terebellidae, Sabellidae, and Maldanidae) present in both locations. We used 24 open-bottom cages to enclose the benthos and its seagrass-associated animal communities. We manipulated each cage by assigning it to one of the following treatments: (1) inclusion of fishes in upper and intermediate trophic levels (1 juvenile gulf flounder and 10 juvenile spot, the flounder + spot treatment); (2) inclusion of the intermediate predator (10 juvenile spot with no gulf flounder, the spot-only treatment); and (3) no fish added (unmanipulated controls). Core samples taken within the cages provided pre- and post-experimental measures of polychaete density and biomass, and the difference in density and biomass were used as response variables. At the end of the experiment, we collected, weighed, and analyzed the gut contents of all juvenile spot present in the cages. Juvenile pinfish ( Lagodon rhomboides, x¯ = 30 mm SL) were present at the end of the study, having arrived as larvae or being trapped during cage set-up, and these fish were also examined, because they also eat polychaetes and their natural densities exceeded our introduced spot densities. Significant differences among treatments were detected for the polychaete family Terebellidae for both the change in density and biomass (pre-experiment − post-experiment). Densities of the Terebellidae changed in the direction predicted by the network model's impact analysis, declining in the cages with spot added compared with the control cages. Analyses of the other response variables (post-experiment spot and pinfish densities and biomass, difference between pre- and post-experiment polychaete densities and biomass for other families, and post-experiment spot and pinfish stomach content biomass) showed no significant differences among treatments. Several variables (Nereidae densities, pinfish densities and biomass, and pinfish stomach content biomass) varied between cages with low and high seagrass cover (significant blocking effect, P < 0.001). Nereidae densities declined significantly in cages with high (73%) rather than with low coverage (31% cover) of seagrass. Pinfish density and biomass were significantly greater in the high seagrass cages at the end of the experiments ( P < 0.001), suggesting that dense seagrass attracted them. We conclude that the high density of pinfish in dense seagrass was responsible for the decline in density of the Nereidae. The direct effect of intermediate predators (pinfish feeding on polychaete prey) can be influenced by preferential recruitment of fishes to structurally complex habitats. The direction of change of indirect effects, but not the magnitude, in multi-trophic-level food webs can be predicted by the mixed trophic impact analysis of network models. However, these indirect effects are likely to be small in magnitude relative to direct effects and may be difficult to detect experimentally, especially in low-power experimental caging studies with natural fluctuations in recruitment rates of competitor species.