This paper reports a numerical instability problem in the widely used Community Multiscale Air Quality (CMAQ) model and discusses its impacts on ozone and particulate matter simulations. By adding 0.5 moles/sec of NO x emissions to Middlesex County, CT, for example, CMAQ (2003 version) predicts up to 1 μg/m 3 change in PM 2.5 concentrations in the Ohio Valley and southern California in less then 48 hours. These regions are beyond the reach of normal transport processes in such a short time, and the remote and upwind responses are 100 times larger than responses near or downwind of the source area. More recently, progress has been made in reducing the numerical instability by correcting coding errors in the transport algorithm, adopting additional vertical wind adjustment to enhance mass conservation, and making numerous improvements in the ISSOROPIA aerosol thermodynamics module (2004 and 2005 CMAQ versions). These improvements, however, are not sufficient to reduce the instability to a reasonable level. The magnitude of peak instability in the 2005 version of CMAQ remains comparable to the normal responses from NO x emissions of a middle-size power plant. This problem, although having a minor effect on the model performance to simulate total O 3 and PM concentrations, results in difficulties when the current version of CMAQ is used to address many important air quality issues including localized emission controls and source-receptor simulations.