Micro- and mesocosms are frequently required in regulatory procedures of aquatic risk assessment for pesticides. However, many questions are still a matter of debate with regard to the use of these systems for environmental risk assessment, especially considering the inter-system variability of the measured parameters and its consequences on experimental design and data analysis. In this paper, variability of physico-chemical and biological parameters measured during two long-term experiments (8 to 9 months) in uncontaminated outdoor freshwater lentic mesocosms (8 m(3)) is analysed. Consequences on the design of ecotoxicity tests in mesocosms and on data analysis are also addressed. Water temperature, pH, dissolved oxygen concentration and concentration of suspended solids exhibited a very low variability whereas nutrient concentrations displayed elevated levels of variability. Among biological parameters, those measured at the individual level were less variable than those measured at the community level. Functional descriptors frequently exhibited a lower inter-mesocosm variability than structural descriptors. Aggregation of data proved to significantly reduce inter-mesocosm variability. The results indicate that univariate statistical methods may be used for physico-chemical or species-level (e.g. biometric parameters) data which exhibit a moderate inter-mesocosm variability. The use of multivariate techniques is suggested for other levels of investigation. Nevertheless, variability is not sufficient to identify useful parameters. The sensitivity towards chemicals and ecological relevance of descriptors within the experimental context must also be considered.