The main requisite for a functional computerized adaptive testing system is the need of a calibrated item bank. This text presents the tasks carried out during the calibration of an item bank for assessing knowledge of Basque language. It has been done in terms of the 3-parameter logistic model provided by the item response theory. Besides, this work focuses on the most important decisions taken and the algorithms followed during the process. Actually, the most significant novelties introduced to the traditional calibration procedure deal with the adaptation of the algorithm and the putting into practice of both supervised and non-supervised web-based subtest administrations. Once the item bank had been calibrated, it was added to the generator of computerized adaptive admission tests that is expected to be operative in the new version of Hezinet, a multimedia Basque language e-learning management system that is being used since 2000 in more than 60 Basque-adult schools. (Contains 7 tables and 4 figures.)