This work tackles the question of automatic generation of differentiated assessment Tests that should guarantee fairness and reasonable difference, by the content and the structure, between all involved learners. More specifically, these researches aim to maximize differentiation in assessment Test Collections, while minimizing the number of necessary Items in the source database. To do so, a metric capable of measuring the differentiation in Test Collections and three generation algorithms are elaborated. Thanks to this metric, the performances of each generation algorithms are measured in terms of differentiation, which allows the generator to choose the most appropriate algorithm depending on the parameters of the teacher at the time of generation. This paper presents some experiments and results based on multiple choice questionnaires, and some perspectives about generation of differentiated Tests.