The use of computational fluid dynamics (CFD) to solve indoor airflow problems has increased tremendously in the last decades. However, the accuracy of CFD simulations depends greatly on user experience, the available validation data, and the effort made to verify solutions. This study presents the results of a conference workshop, which assessed user influence on the CFD results obtained for a generic non-isothermal flow problem; ie, a backward-facing step flow problem with a heated wall below the supply. Fifty-five simulation sets were submitted by 32 teams. The results showed a very large spread in predicted penetration length (xre/(H − h)), location of maximum velocity in the lower part of the recirculation cell (xrm/(H − h)), and maximum velocity at this location (urm/u0). The turbulence model seemed to very strongly influence the results, with a statistically significant difference in the predictions yielded by the k-ε and k-ω models. The results obtained using a single turbulence model generally also showed a spread in results; the level of spread depended on factors such as grid size and near-wall treatment. The statistical data strongly indicate the need for validation studies using experimental data (benchmarks) to ensure the accuracy, reliability, and trustworthiness of CFD simulations for indoor airflow problems.