This article summarizes the first part of a benchmark exercise comparing seven different commercial three dimensional codes used for dispersion modeling in the context of major accident risk assessments. The aim of this first stage was twofold: First, the project intended to better understand the potential magnitude of the variation in results for a relatively simple study case. The benchmark case confirmed that, just as with integral models, the use of different codes can cause significant variation in the dispersion results. Second, the study wanted to identify the key assumptions which had a predominant influence on the uncertainty of the results, with the intention of providing clear guidance in engineering specifications for the definition of the scope of work of a computational fluid dynamics dispersion study. This article summarizes the key drivers identified, to date, for the variation in results, and the work shows this variation can be reduced to an acceptable margin by clearly specifying these key assumptions. (c) 2013 American Institute of Chemical Engineers Process Saf Prog, 2013