Emission inventories are generally identified as the key input to the air quality modelling chain, especially when they are used to support regulatory decisions, such as for air quality planning or for the assessment of concentration levels over a given territory. At the same time, studies point out to emission inventories as the most uncertain factor among the different components of air quality models. In a recent work, Thunis et al. (2016), developed a methodology, supported by a specific screening diagram, to identify discrepancies between emission estimates and target the pollutants and sectors for which improvements should be prioritized. Based only on the total emissions for various pollutants as input, the methodology is able to provide insight on whether these differences arise from issues related to emission factors or activities. In this work we further develop this methodology and show that the use of a probabilistic approach improves its usefulness and relevance. We motivate the use of a probabilistic approach by discussing a series of simple situations to which we apply an “intuitive reasoning”. These situations are then used as background to detail the probabilistic methodology and its main assumptions. Tested on a random set of known emission inventories, we show that the methodology performs well in reproducing the expected activities and the associated emission factors. We show that the method becomes more precise when the number of pollutants increases. Given the large differences observed between emission inventories, reducing the discrepancies between them does not only lead to more coherence but it also improves their accuracy as errors can be detected and solved. The approach is mostly designed as a screening to spot the main inconsistencies in the field of atmospheric emissions but the methodology is general and could be applied to other fields, provided that the relationships between variables fulfil similar rules as those described here.