This study aims to investigate the predictive validities of three different traffic psychological assessment systems used in Turkey (ART 2020, TRAFIKENT and Vienna Test System) by using systematic observation of driving behaviour in natural traffic settings. For this purpose, 176 drivers were evaluated by observers and then tested in three different traffic psychological assessment systems. Drivers participated in a 45 km (approx. 80 min) driving session consisting of diverse road types. Observers filled systematic observation forms comprising 35 errors/violations on 10 road segments. Furthermore, observers evaluated participants regarding cognitive and psychomotor abilities measured with the psychological assessment system. Consequently, although the systems' power to predict on-road behaviour was similar, the tests/abilities that contributed to the models differed in each system except for the reasoning test. It was recommended to increase the weight of these predictive tests in deciding safe driving for each system.