Relationships between surface sediment diatom assemblages and measured environmental variables from 77 lakes in the central Canadian arctic treeline region were examined using multivariate statistical methods. Lakes were distributed across the arctic treeline from boreal forest to arctic tundra ecozones, along steep climatic and environmental gradients. Forward selection in canonical correspondence analysis determined that dissolved inorganic carbon (DIC), dissolved organic carbon (DOC), total nitrogen (TN), lake surface area, silica, lake-water depth, and iron explained significant portions of diatom species variation. Weighted-averaging (WA) regression and calibration techniques were used to develop inference models for DIC, DOC, and TN from the estimated optima of the diatom taxa to these environmental variables. Simple WA models with classical deshrinking produced models with the strongest predictive abilities for all three variables based on the bootstrapped root mean squared errors of prediction (RMSEP). WA partial least squares showed little improvement over the simpler WA models as judged by the jackknifed RMSEP. These models suggest that it is possible to infer trends in DIC, DOC, and TN from fossil diatom assemblages from suitably chosen lakes in the central Canadian arctic treeline region.