Most general models do not include herbivory as a major agent of successional change. Potentially, herbivores can affect succession in three ways: accelerating or decelerating the rate of succession, where the sequence of dominant species is unaltered, or deflecting succession onto a new trajectory, where the species composition of dominants becomes substantially different than during ungrazed succession. We examined these alternatives for benthic algae on a coral—reef crest off Oahu, Hawaii. In this system, exposed coral—rock surfaces naturally undergo one of two major grazing regimes: (1) relatively protected inside defended territories of the damselfish Stegastes fasciolatus (Pomacentridae), where the benthos is dominated by filamentous algae; or (2) exposed to abundant schooling parrotfishes (Scaridae) and surgeonfishes (Acanthuridae) outside territories, where the bottom is covered mostly by crustose algae. We compared the effects of this differential grazing on primary succession, relative to ungrazed succession, by distributing on the same date 1332 settling surfaces among three treatments: exposed inside damselfish territories, exposed just outside territories, and within fish—exclusion cages just outside territories. To balance the advantages and disadvantages of different settling surfaces, we used equal numbers of each of three kinds of 50—cm2settling plates: naturally contoured coral rock, coral rock cut into flat plates, and roughly sanded PVC plastic. To follow relative successional pathways, we sampled destructively 63 plates (21 from each grazing treatment) 17 times over 1 yr. Plates placed in the field several months before and after the main experiment suggested no seasonal differences in algal colonization. A concurrent cage—control experiment involving 144 settling plates, combined with measurements of light and water motion inside vs. outside cages, indicated that the secondary effects of cages were minor compared to the primary effect of preventing fish grazing. In the absence of fish grazing within cages, algal succession over the year followed three stages: early dominance by simple green and brown filaments (such as Enteromorpha rhizoidea and Ectocarpus indicus), a midsuccessional stage dominated by thin and finely branched red filaments (such as Centroceras clavulatum and Taenioma perpusillum), and a late stage dominated by blades and coarsely branched thick filaments (especially Tolypoicladia glomerulata). Species diversity followed a unimodal pattern during ungrazed succession, declining as a few species of late—stage algae predominated. Inside damselfish territories, succession was decelerated. The early stage was protracted and the midsuccessional stage, similar to natural assemblages inside territories, still dominated by the end of the year. Here, herbivory was of moderately destructive intensity (as measured by the density of fish bite marks that removed algal holdfasts) and fairly nonselective (as measured by comparisons of the gut contents of damselfish paired with samples of their algal mats). Algal biomass reached only about a quarter of what accumulated during ungrazed succession, but species diversity gradually increased through time. By the end of the experiment, algal species diversity was greatest inside damselfish territories compared to the other two grazing treatments. Outside territories, where grazing was destructively intense, resulting in the removal of all erect algae, succession was strongly deflected. The early stage was quickly replaced by a low—biomass and low—diversity assemblage of crusts (such as Hydrolithon reinboldii) and prostrate blue—green mats (such as Calothrix crustacea), characteristic of natural assemblages outside territories. Besides demonstrating the importance of herbivory during succession and providing insight on the mechanisms involved, these patterns have ramifications for explaining the maintenance of high local species diversity on coral reefs at two spatial scales. Between patches, differential grazing by territorial damselfish vs. schooling herbivores causes succession to follow different trajectories toward different algal assemblages. Within patches defined by damselfish territories, moderate grazing decelerates succession and prolongs a high—diversity mid—successional stage. Both these patterns provide an example of predation maintaining high local diversity in tropical systems, and indicate that territorial damselfish can function as keystone species on coral reefs.