A new network of 21 Picea engelmannii Parry ex Engelm. ring-width chronologies was developed from tree-line sites in the Canadian Rockies. These chronologies range in length from 297 to 648 years (mean 423 years) and have mean sensitivities between 0.16 and 0.20 (mean 0.18), first-order autocorrelations between 0.73 and 0.88 (mean 0.83), and subsample signal strengths >0.85 for 246-494 years (mean 324 years). Mean intersite correlations (A.D. 1700-1982) for chronologies with expressed population signals >0.85 are 0.46 and 0.63 for standard and residual chronologies, respectively. Standard ring-width chronologies are, in general, positively correlated with summer temperatures and negatively correlated with spring and previous summer temperatures. A regional June-September temperature reconstruction for the Banff-Jasper region (BJR; A.D. 1715-1982) was developed using multiple regression of three significant principal components from 14 standard ring-width chronologies. The first principal component contains 55% of the total chronology variance. The model reconstructs 38% of summer temperature variance during the calibration period (1888-1982). The BJR is the first regional temperature reconstruction for this area based on ring-width data from a network of sites. The reconstructed temperature patterns are broadly similar to other regional estimates of past temperatures. Above-average summer temperatures occurred in the mid-20th century and the late 1700s - early 1800s. Most of the 19th century was unusually cold, with the coldest conditions in the late 19th century. Detailed differences between BJR and previously developed reconstructions lie well within 2 sigma confidence limits and may reflect differences in tree species, modelling techniques, spatial coverage, and the seasonal temperature parameter reconstructed.