We assessed impacts of the 'Exxon Valdez' oil spill on bentluc communities within and adjacent to eelgrass beds in Prince William Sound, Alaska, USA. The concentration of total polynuclear aromatic hydrocarbons (TPAHs),'benthic community composition. diversity, biomass, and abundance were compared between matched pairs of oiled and reference sites in 1990 (approx. 16 mo after the spill), andin 1991, 1993, and 1995. TPAHs in sediments were high (up to 15300 ng g.') at sites adjacent to oiled shorelines in 1990, but declined sharply thereafter. Some reference sites in 1990-91 also had elevated TPAHs in sediments and signatures matching Exxon Valdez oii, but concentrations were significantly lower than at oiled sites. Based on classification and ordination analyses, cornrnunities of infauna and small epifauna at some oiled sites in 1990 differed from communities at reference sites, and from the Same sites in subsequent years. Percent sand and mud aild concentration of total chrysenes (PAH analytes indicative of crude oil) explained significant proportions of the temporal and spatial variation in benthic community structure. Total abundance and biomass of epifauna were generaliy higher at oiled sites, primarily because of higher densities of epifaunal bivalves. Otherwise, there were few consistent cornrnunity-wide responses to oiling in diversity. richness, total abundance. total biomass. or the abundances of major taxonomic groups (e.g. polychaetes or bivalves). We attnbute the lack of a stronger community-wide response to the varying sensitivities of constituent taxa to oil and organic enrichment. Over half of the dominant famiiies differed with respect to abundance at oiled versus reference sites Most, including 9 families of polychaetes, were more abundant at oiled sites. Most of these were stress-tolerant or opportunistic. and their increase was likely due to organic enrichment. Negative impacts were most evident in oil-sensitive amphipods, especially the families Isaeidae and Phoxocephalidae. There were consistently more of these amphipods at reference sites, and abundances at oiled sites were likely reduced as a result of oil toxicity. Most of these differences between oiled and reference sites persisted through 1995,6 yr after the spill. We suspect that these differences are a result of the spill, but we rely on post-spill coinparisons to infer impacts, and our conclusions rely on the untestable assumption of equality between oiled and reference sites in the absence of a spill. Future assessments of the impacts of oil spiiis or other accidental environmental disturbances could benefit from pre-iinpact studies that provide objective critena for selection of matched pairs of sites, thereby supporting the assumption of equality in the absence of the disturbance.