Estimating ecosystem-level effects from single-species bioassays is a major challenge in environmental risk assessment. Most extrapolation procedures are based on the implicit assumption that species sensitivities are random with regard to their functional traits. Here, we explore how nonrandomness in species sensitivities affects how species-level and ecosystem level effects of chemical exposure correspond. The effect of a correlation between the trait value under control conditions and the sensitivity of the trait to chemical stress is studied for two traits (per capita growth rate and monoculture yield) under constant and temporary exposure. Theoretical model predictions are thereby validated against a 3-week microcosm experiment, in which eight marine diatoms systems with different correlations between trait values and sensitivities were temporary (1 week) or constantly (3 weeks) exposed to two concentrations of the herbicide atrazine (100 and 250 μg L-1). Negative correlations increased the reduction in ecosystem functioning (productivity) by atrazine for both traits. However, correlations in the per capita growth rate affected productivity only shortly following changes in environmental conditions, whereas correlations in the monoculture yield affected productivity throughout exposure. Correlations between species sensitivities and functional trait values can thus help to identify when ecosystem-level effects are likely to exceed species-level effects.