Abstract: Many Intelligent Tutoring Systems (ITS) using a learning machine companion called Learning Companion Systems (LCS) has been developed these last years. These systems involve three agents: a teacher, a student and a companion. These systems use the idea that the social interaction has an influence on the cognitive development. The basic idea is that the student can learn from his mistakes and can benefit from the companion's mistakes. The companion is looked for assistance, and other times as a troublemaker. The goal of the learning companion is to stimulate through collaboration competition and demonstration. This paper outlines the architecture of an intelligent tutoring system based on the Double Test Learning (DTL) strategy and using two companions as a co-students where one of both is a troublemaker. The system called DB-TUTOR is proposed to teaching the databases for the students of the third year of the graduate cycle. The DTL strategy introduces a co-student in the learning session that receives the same training as the human student. The idea of the DTL strategy, which contains four phases, is based on that the student will benefit from the co-students's mistakes and their behavior. When the training phase is completed, the tutor will test the co-students. In this phase called Post-Test1, the student will follow the interaction between the tutor and the co-students and he can mention observations about the behavior of the co-students in a notebook that is created for that fact. This interaction consists of a sequence of question/answer that is evaluated by the tutor. In the second phase called post-test2, the tutor will evaluate the human student, which will react to the interaction without his notebook. We propose a student model taking into account the DTL strategy for the evaluation phase. A prototype of the DB-TUTOR system has been developed under DELPHI 5 environment.