The successful computer simulation of physical processes consists of two complementary stages: (1) the development of a numerical model that represents the physical process and (2) its evaluation. Although less often discussed, the second stage is as important as the first one. In the second stage, the goal is to verify as much as possible that the model cannot only reproduce the observations but that it does it for the right reasons. Techniques that take that factor into account are presented in the present paper. They include the use of spectral and cross-spectral analysis and the fitting of time series models to the observed and simulated data. The time series models used are composed of five components describing the long-term, episodic, diurnal variations as well as an autoregressive and a random component. To evaluate the numerical model, the components of the observed and simulated values within the same grid cell are compared. The different techniques described in the paper are illustrated by evaluating the chemical tracer model (CTM) model described by Pudykiewicz et al. <1997>. The model was run for the period between June 15, 1995, 0000 UTC and August 16, 1995, 0000 UTC. Hourly mean O3 mixing ratios were generated for each cell within the model domain. The numerical results were compared to data at 322 sites across the model domain from the NARSTO-NE database. The paper also presents suggested modifications to the CTM model to improve it. ¿ 1999 American Geophysical Union