Several occupational exposure models are recommended under the EU's REACH legislation. Due to limited availability of high-quality exposure data, their validation is an ongoing process. It was shown, however, that different models may calculate significantly different estimates and thus lead to potentially dangerous conclusions about chemical risk. In this paper, the between-model translation rules defined in TREXMO were used to generate 319000 different in silico exposure situations in ART, Stoffenmanager, and ECETOC TRA v3. The three models' estimates were computed and the correlation and consistency between them were investigated. The best correlated pair was Stoffenmanager-ART (R, 0.52-0.90), whereas the ART-TRA and Stoffenmanager-TRA correlations were either lower (R, 0.36-0.69) or no correlation was found. Consistency varied significantly according to different exposure types (e.g. vapour versus dust) or settings (near-field versus far-field and indoors versus outdoors). The percentages of generated situations for which estimates differed by more than a factor of 100 ranged from 14 to 97%, 37 to 99%, and 1 to 68% for Stoffenmanager-ART, TRA-ART, and TRA-Stoffenmanager, respectively. Overall, the models were more consistent for vapours than for dusts and solids, near-fields than for far-fields, and indoor than for outdoor exposure. Multiple linear regression analyses evidenced the relationship between the models' parameters and the relative differences between the models' predictions. The relative difference can be used to estimate the consistency between the models. Furthermore, the study showed that the tiered approach is not generally applicable to all exposure situations. These findings emphasize the need for a multiple-model approach to assessing critical exposure scenarios under REACH. Moreover, in combination with occupational exposure measurements, they might also be used for future studies to improve prediction accuracy.