This paper reports on the performance of Atmospheric Dispersion Modelling System (ADMS) 4.2 in predicting peak and mean ambient sulfur dioxide concentrations at two sites adjacent to the Map Ta Phut Industrial Estate in Eastern Thailand, the centre of the country’s petrochemical industry. The model comprised 100 individual stacks and utilised four separate meteorological datasets from different points around the site. We show that model performance varies according to the location at which the meteorological data were obtained, with considerable differences in model outputs observed for meteorological stations that are relatively close to each other. The best performances were observed when there was co-location of the meteorological data and receptor. In such cases, acceptance criteria for the majority of performance parameters were satisfied across averaging periods ranging from 1 h to 7 days. We have also compared the results from this study with those obtained from a recent literature American Meteorological Society/United States Environmental Protection Agency Regulatory Model (AERMOD) study for the same site and time period; the comparison indicates that AERMOD is likely to be similarly influenced by the choice of meteorological dataset. Using ADMS model simulations for all four meteorological datasets and a breakdown of the local population by electoral ward, we were able to estimate exposure over 1 h, 24 h and yearly averaging periods and compare these to air quality standards and guidelines published by Thailand, the World Health Organisation (WHO) and the European Union (EU). The results of this analysis showed that despite the large variations in overall model performance, the impact of choice of meteorological dataset on prediction of compliance with the standards and guidelines is relatively small: the WHO 24-h guideline of 7.5 ppb (100th percentile) was predicted to be exceeded in all of the wards for all meteorological datasets, whilst compliance with Thai and EU standards was predicted for at least 86 % of the population, with relatively little variation between the different meteorological datasets.