Behavior evaluation is an approach to the problem of the detection of intruders that are undesirable in the computer system. Considering multi–agent architecture each agent should execute continuous and autonomous behavior evaluation of other agents existing in the environment of the secured system. This means that an agent is evaluated separately by all agents in the environment. The distributed character of behavior evaluation in multi–agent system requires an algorithm of management and collection of the results of autonomous behavior evaluations of agents. The algorithm of results’ collection should enable to elect the worst agent or agents which have to be eliminated. The main topic of this article is a modified approach to the behavior evaluation process. This approach reduces the number of evaluations which have to be done by agents. The main idea is to evaluate only sampled actions (e.g. every second action) which are undertaken by agents in the secured system.