Effective and accurate modeling of air quality in complex terrain constitutes one of the main challenges for the community of modelers. One of the basic problems is the selection of input data of adequate quality in combination with a uniform configuration of the modeling systems. Simultaneously, the primary aim is to obtain predictions of satisfying accuracy. In the article the results of the evaluation for the CALMET/CALPUFF modeling system in near-field and complex terrain were presented. Research was conducted based on three experimental databases for the dispersion model evaluation, i.e. Martin's Creek (SO2), Lovett (SO2) and Tracy Power Plant (SF6). Each experiment concerned an area characterized by different topography, meteorological conditions, emission source features and the location of tracer substance monitors (SO2, SF6). The aim of the study was to determine the optimal settings for the CALMET/CALPUFF models with regard to the digital elevation model dataset (GTOPO30, SRTM3, NED), grid resolution (ranging from 100 to 4000 m) and the terrain adjustment methods available in the CALPUFF model (MCTADJ = 0, 1, 2 and 3). The results of the CALMET/CALPUFF accuracy evaluation showed, that the use of the digital elevation model (DEM) with a horizontal accuracy of approximately 90 m and a vertical accuracy of 15 m (SRTM3) is sufficient. Application of more accurate data (NED) resulted in comparable model evaluation outcomes. Using both dense and coarse grids resulted in FB and/or FBRHC higher than 0.6. Omitting the terrain adjustment method in the CALPUFF model results in the twofold underestimation of the measurements. The best results of the CALMET/CALPUFF accuracy evaluation for each experiment (|FB| < 0.30, FAC = 1.0, NAD ≤ 0.15, VG < 1.12, IOA > 0.5) were achieved for the grid resolution of 400 m with the use of a partial plume path adjustment method (MCTADJ = 3).