The classical approach for network diagnosis systems is based on centralized solutions with deterministic algorithms. However, the increased complexity of new networks and services, including customers premises, demands new strategies to face two key challenges: scalability of the system and reliability against incomplete or inaccurate information. This paper describes KOWLAN, a Multiagent system in which agents carry out Bayesian inference to drive the diagnosis process and to obtain diagnosis results. The system has been quickly developed and successfully deployed, and it is contributing to a reduction of the attendance time.