Marine protected areas (MPAs) are increasingly envisaged as a tool to manage coastal ecosystems and fisheries. Assessment of their performance with respect to management objectives is therefore important. A number of WAS provided conservation benefits for fished species. Observed benefits do not apply to all species at all times, and responses to protection are also highly variable among fish taxa. Among the many empirical studies on marine reserves, only a few designs considered 'before and after data' and spatial variation. In this paper, we are interested in assessing the effect of a no-take reserve on the reef fish assemblage in a northwestern Mediterranean example. Data were obtained from a three-year survey using underwater visual censuses (UVC), before and after MPA establishment. Permutational multivariate analysis of variance (PERMANOVA) and multivariate regression trees (MRT) were used to evaluate the effects of reserve protection on the reef fish assemblage, while accounting for habitat. Modelled biological responses were abundances and diversity indices calculated at different levels of the assemblage. Significant effects were found for many of these metrics. in addition to PERMANOVA, univariate models provided more insight into the magnitude and direction of effects. The most sensitive metrics were related to large species and species targeted by fishing. These results may be used to choose the metrics that are more suitable as community-based indicators of MPA impact in the perspective of monitoring programs. (c) 2006 Elsevier Ltd. All rights reserved.