Abstract: The article concerns the computer testing system architecture, possessing the elements of intelligence. The intelligence of the testing system involves adaptability of the testing process, as well as the ability to diagnose and identify errors in the decision process, and it is provided for by the use of the knowledge base of the subject area, the structure of the complexity of case tasks, as well as the expert subsystem and the competency model of students. The subject of the study concerns method and results of implementing an automated assessment of the level of formed subject competencies among the students based upon an analysis of their task decisions generated in accordance with the structure of the level of complication of typical tasks. The options for the use of the intellectual testing system in the context of online learning involve the options of the component architecture of the implemented system as learning subsystems embedded in LMS Moodle, diagrams of the system functioning and usage. The research involves the analysis of the implemented environment architecture for solving learning tasks, as well as and the structure of decision-making concerning the level of the formed subject competencies of the students. The results of introducing a subsystem into a real online course are analyzed. The main results of the study are as follows: the architecture of the subsystem used is sufficiently flexible and makes it possible to provide the registration and logging of decisions for users of the testing system, provides the freedom of the student in the construction of possible solutions; comparative analysis of the solution with the decisions of the agent-experts makes it possible to evaluate the rationality of the solutions; construction of student competency map as overlay model makes it possible to assess the level of the formed subject competencies in accordance with the level of complexity of tasks; it allows organizing adaptive testing that takes into account the individual characteristics of students; the system is open for further development through the introduction of multi-agent and neural network solutions.