Regional source attribution is conducted for fine particulate sulfate at Big Bend National Park, Texas, using a comprehensive regional air quality model, Community Multiscale Air Quality model augmented with the Model of Aerosol Dynamics, Reaction, Ionization and Dissolution (CMAQ-MADRID), as part of the Big Bend Regional Aerosol Visibility and Observational (BRAVO) Study. The overall PM2.5 sulfate load at Big Bend National Park over the 9 July to 28 October 1999 period is attributed as follows: 31% to Mexico, 19% to Texas, 39% to the eastern United States, 6% to the western United States, and 5% to areas outside the modeling domain (boundary conditions). The arithmetic mean of the daily PM2.5 sulfate loads at BBNP over the 9 July to 28 October 1999 period is attributed as follows: 42% to Mexico, 14% to Texas, 27% to the eastern United States, 9% to the western United States, and 7% to areas outside the modeling domain. These results illustrate the potential for significant contributions from distant sources to regional haze in remote areas. An examination of source contributions and model performance by month and for specific episodes shows that model performance can affect the results of a source attribution. Therefore caution is advised when interpreting the results of source attribution obtained using Eulerian air quality models. In lieu of estimating the uncertainty of the apportionment procedures, source region attribution results obtained for sulfate using CMAQ-MADRID are refined using an inverse modeling technique. Comparison of original attribution results with refined attribution estimates obtained using inverse modeling techniques shows that these methods can reduce in part the biases introduced in the model by uncertainties and errors in the emissions, meteorology, and chemical transport modeling. The refined attribution estimates of overall PM2.5 sulfate load at Big Bend National Park over the 9 July to 28 October 1999 period using inverse modeling are 37% to Mexico, 17% to Texas, 31% to the eastern United States, 9% to the western United States, and 6% to areas outside the modeling domain.