Regional or local climate change modeling studies currently require starting with a global climate model, then downscaling to the region of interest. How should global models be chosen for such studies, and what effect do such choices have? This question is addressed in the context of a regional climate detection and attribution (D&A) study of January-February-March (JFM) temperature over the western U.S. Models are often selected for a regional D&A analysis based on the quality of the simulated regional climate. Accordingly, 42 performance metrics based on seasonal temperature and precipitation, the El Nino/Southern Oscillation (ENSO), and the Pacific Decadal Oscillation are constructed and applied to 21 global models. However, no strong relationship is found between the score of the models on the metrics and results of the D&A analysis. Instead, the importance of having ensembles of runs with enough realizations to reduce the effects of natural internal climate variability is emphasized. Also, the superiority of the multimodel ensemble average (MM) to any 1 individual model, already found in global studies examining the mean climate, is true in this regional study that includes measures of variability as well. Evidence is shown that this superiority is largely caused by the cancellation of offsetting errors in the individual global models. Results with both the MM and models picked randomly confirm the original D&A results of anthropogenically forced JFM temperature changes in the western U.S. Future projections of temperature do not depend on model performance until the 2080s, after which the better performing models show warmer temperatures.