The density, biomass and estimated production of brook trout Salvelinus fontinalis and Atlantic salmon Salmo salar were related to habitat factors in streams of Terra Nova National Park, Newfoundland, Canada. Fish communities at 29 sites (18 brooks; 15 watersheds) were sampled in the summer of 2002, 2003 and 2005. Salmonid density, biomass per unit area and production (derived from biomass and fish size using allometric P:B relationships) were compared with site habitat characteristics (wetted width, lactustrine habitat, per cent riffle habitat, canopy coverage and stream gradient), using an interactive stepwise multiple linear regression. Salmonid biomass (mean: 2·87 g m−2; range: 0·33–10·88 g m−2) and estimated production (mean: 3·05 g m−2 year−1; range: 0·32–10·98 g m−2 year−1) within the study area varied by an order of magnitude, however, habitat variables accounted for much of this variation. Specifically, wetted width and lacustrine area of the tributary played important roles in explaining density, biomass and production. Wetted width was important for all measurements of brook trout and total salmonids while lacustrine area was important for all measurements of Atlantic salmon and played a lesser role in total salmonid biomass. Other factors such as the percentage of riffle habitat, site gradient and canopy coverage provided modest improvements to the fit of some relationships. When models using the same environmental factors were compared, those using production estimates derived from allometric P:B equations in the literature provided improved predictive capability than did those from direct density and biomass estimates. It is proposed that allometric P:B relationships have utility in improving comparisons of stream fish communities, particularly in studies with insufficient resources to measure production directly.