The use of biomarkers to assess the impacts of contaminants on aquatic ecosystems has noticeably increased over the past few years. Few of these studies, however, have contributed to the prediction of ecologically significant effects (i.e., at the population or community levels). The present field study was designed to evaluate the potential of metallothionein (MT) and sub-cellular metal partitioning measurements for predicting toxic effects at higher levels of the biological organization in freshwater bivalves (Pyganodon grandis) chronically exposed to Cd. For that purpose, we quantitatively sampled P. grandis populations in the littoral zone of nine lakes on the Precambrian Canadian Shield during two consecutive summers (1998 and 1999); lakes were characterized by contrasting Cd levels but similar trophic status. We tested relationships between the population status of P. grandis (i.e., growth parameters, density, biomass, secondary production, turnover ratio and cumulative fecundity) and (i) ambient Cd concentrations, (ii) sub-organismal responses (MT concentrations in the gill cytosol of individuals and Cd concentrations in three metal-ligand pools identified as M-HMW, the high molecular weight pool, M-MT, the metallothionein-like pool and M-LMW, the low molecular weight pool) and (iii) ecological confounding factors (food resources, presence of host fishes for the obligatory parasitic larval stage of P. grandis). Our results show that littoral density, live weight, dry viscera biomass, production and cumulative fecundity decreased with increasing concentrations of the free-cadmium ion in the environment (Pearson's r ranging from -0.63 to -0.78). On the other hand, theoretical maximum shell lengths (L( infinity )) in our populations were related to both the dissolved Ca concentration and food quality (sestonic C and N concentrations). Overall, Cd concentrations in the gill cytosolic HMW pool of the individual molluscs were the biomarker response that was most frequently and most strongly correlated with the population variables (Pearson's r ranging from -0.58 to -0.80). Our findings demonstrate, however, the difficulty of currently assigning to sub-cellular metal partitioning measurements (mainly Cd bound to the HMW fraction) any predictive role for population health, notably because of the influence of ecological confounding variables (e.g., the cumulative number of degree-days in the littoral zone, as is the case here). Metal contamination of our lakes has decreased markedly in the past 10 years and consequently we believe that the toxic effects of metals may have been replaced by some natural factors as the main agent for structuring the clam populations in these lakes.