Experimental ecosystems such as mesocosmshave been developed to improve the ecological relevanceof ecotoxicity test. However, in mesocosm studies, thenumber of replicates is limited by practical and financialconstraints. In addition, high levels of biological organizationare characterized by a high variability of descriptivevariables. This variability and the poor number of replicateshave been recognized as a major drawback fordetecting significant effects of chemicals in mesocosmstudies. In this context, a tool able to predict preciselycontrol mesocosms outputs, to which endpoints in mesocosmsexposed to chemicals could be compared shouldconstitute a substantial improvement. We evaluated here asolution which consists in stochastic modelling of thecontrol fish populations to assess the probabilistic distributionsof population endpoints. An individual-basedapproach was selected, because it generates realistic fishlength distributions and accounts for both individual andenvironmental sources of variability. This strategy wasapplied to mosquitofish (Gambusia holbrooki) populationsmonitored in lentic mesocosms. We chose the number offounders as a so-called ''stressor'' because subsequentconsequences at the population level could be expected.Using this strategy, we were able to detect more significantand biologically relevant perturbations than using classicalmethods. We conclude that designing an individual-basedmodel is very promising for improving mesocosm dataanalysis. This methodology is currently being applied toecotoxicological issues.