The general trend toward probabilistic risk assessment in aquatic toxicology relies on some key assumptions, several of which are questionable. At present, only a few studies have been carried out to evaluate the probabilistic approach. To this end, a multi-species outdoor microcosm study was conducted using the organophosphorus insecticide azinphos-methyl, the freshwater fish ' Pimephales promelas' and 'Lepomis gibbosus', as well as zooplankton. A regression design was used with 5 treatment levels (0.1-100 [mu]g/L) and a control. Comparison of the 96 hr LC50s for fish (31.7 and 3.2 [mu]g/L, respectively) with the 10th percentile of the acute sensitivity distribution (0.9 [mu]g/L) indicates that this extrapolated value would not correspond to acute toxicity in these species. Based on literature data, the multi-species no observable effect concentration (NOEC) for ' Daphnia' sp. (0.1 [mu]g/L) coincides with the chronic and acute 10 th centiles (0.092 and 0.15 [mu]g/L) of the single species invertebrate sensitivity distributions. For fish, the multi-species NOEC for survival (0. 17 [mu]g/L is below the acute 10th centile of 0.9 [mu]g/L, which is not protective. Probabilistic risk assessment of azinphos-methyl in Southern Ontario surface waters indicates that up to 30% of fish and 50-70% of invertebrate species may be acutely impacted. The applicability of probabilistic extrapolation to ecosystem level effects is highly uncertain and thus difficult to reconcile.