Exploratory analysis of field-derived data highlighted a number of important statistical and experimental design considerations for aquatic mesocosm studies. Abundance data for numerically dominant zooplankton taxa were characterized by distinct heterogeneity of variance (coefficient of variation range 10 to 173%) proportional to the mean and related to both time of observation (0 to 77 days postapplication) and taxonomic level of investigation. Log-transformation did not obviate violations in parametric statistical assumptions, necessitating the use of procedures robust to or not assuming variance homogeneity. Using appropriate RM-ANOVA techniques and multiple comparison tests, an overall LOEC value of 1.0 ppm was estimated for effects on numerical abundance. Log-transformation served to linearize the inherently curvelinear concentration-response relationship. Threshold and temporal effects on the concentration-response relationship resulted in significant lack of fit for the linear model in many cases. Statistically significant, adequately fitting models were observed in 21 of 63 cases, with 16 of these meeting the minimum criteria (Fregression/F(0.05;1,12) > 4) for predictive value of the model. A nonlinear regression technique using an exponential decline model wherein both a and b parameters were considered as linear functions of time provided good fit statistics for 5/8 taxa investigated. Examination of residuals provided a posteriori evidence that the nonlinear model adequately addressed the three-dimensional nature of the problem. EC50 values estimated using linear and nonlinear regression procedures differed markedly, ranging from 0.004 to 0.045 and 0.06 to 1.26, respectively. Potential improvements to experimental design and data analysis, including disproportionate replication and three-dimensional methods of data analysis, were identified in relation to regulatory assessment of aquatic mesocosm impact data.