The Swedish driver's license test is a criterion-referenced test with a highly specified curriculum. The test consists of six versions intended to be parallel regarding content and level of difficulty. Today the structure of each test version follows a theoretical model and this model was compared with a new theoretical model. The main purpose of this paper was to analyse if the test versions were parallel by means of theoretical and empirical analyse. The empirical analysis was made with statistical measurement techniques, such as percentage correct, mean, coefficient alpha, and rank correlation coefficient. The results showed that the six test versions were not parallel from the point of view of content, when our new theoretical model was used. As for the level of difficulty the statistical measurement techniques also showed that the six versions of the driver's license tests were not parallel. Gender differences were in favour of women. As a consequence of the results some advice is given for evaluating items as well as complete versions of the driver's license test, for example to use theoretical as well as empirical models, to use try-out data, to change the model for standard setting and to use different kidns of reference groups.