This paper presents, to our knowledge, the first intercomparison and evaluation of complex terrain simulations using three mesoscale models (RAMS, MM5, and Meso Eta) at a horizontal resolution finer than 1 km. Surface and upper-air data from the October 2000 VTMX field campaign in the Salt Lake Valley were used to evaluate the models. The comparison and evaluation focused on boundary-layer structure and evolution and terrain-induced flows in the valley. All three models produced a cold bias from the surface to the top of the valley atmosphere. The models also under-estimated the strengths of the nocturnal inversion, especially over the valley floor where a cold air pool formed at night. RAMS and MM5 captured the development of valley and canyon flows reasonably well, while Meso Eta significantly under-predicted these terrain-induced flows and the associated convergence and divergence in the valley. Overall, RAMS and MM5 outperformed Meso Eta. However, the performance of Meso Eta exceeded our expectations given that the Meso Eta employed one grid compared to the five nested grids in MM5 and RAMS and this was the first time Meso Eta has been run at such fine scale. The study shows that despite the high horizontal resolution that permitted small-scale features to be resolved, relatively large forecast errors can still exist. Therefore, accurate mesoscale forecasting in areas of complex terrain depends on the development of improved parameterizations of surface and boundary layer processes.