We analyzed environmental relationships of the Siskiyou Mountains salamander, comparing attributes at the landscape, macro- and micro-environmental scales, and the three scales combined, to determine which attributes best predicted salamander presence. Separate analyses were conducted for sites on the north and south sides of the Siskiyou Mountains which basically divide Oregon and California. We sampled 239 randomly selected sites (163 north and 76 south of the Siskiyou Crest), each in ≥ 5 ha of relatively homogeneous forest or post-forest (clearcut) habitat, 75 m from edge, in 7×7 m plots with ≥ 25 % rock cover. Measured attributes at the landscape, macro-, and micro-environmental scales totaled 230 independent variables. Subsets of 122 (North Slope) and 97 (South Slope) variables were used in hierarchical, exploratory, best subsets logistic regression, to determine the best predictors of salamander presence. We tested three analysis strategies: 1) linear relationships only; 2) linear, quadratic, and logistic relationships; and 3) linear, quadratic, and logistic relationships, and interactions among the covariates. The best models of salamander presence consisted of combined landscape, macro- and micro-environmental scale variables; included linear, quadratic, and pseudo-threshold (i.e., log) forms, and included interactions between variables. These models showed positive relationships of salamander presence with site conditions and plant assemblages characterizing old, less disturbed forest with closed canopy, moist, relatively warm microclimates, deep litter, and cobble and boulder-sized rock substrates. Our results suggest that mature to late-seral-forest attributes provide optimal habitat for the Siskiyou Mountains salamander. Stands of mature and older forests evenly distributed and interconnected across the geographical range of this species would likely best insure its long-term viability.