The Internet and most current intranet networks are experiencing a huge increase in the volume of traffic. This affects directly the network congestion by saturating the buffers at the routers and contributes to generating lots of data losses as well as reception and transmission delays. The existing TCP end-to-end congestion control uses Additive Increase Multiplicative Decrease (AIMD) approach, a time out and slow start behavior, which lead to data throughput with abrupt changes. Therefore, developing new congestion control strategies based on non-analytical approaches will certainly help to overcome the current difficulties of the internet in particular which are due to network structural complexity, diversity of services supported, and to variety of parameters involved. This work presents a fuzzy logic-based approach for controlling the network congestion. Its main objective is to optimize the available bandwidth and keep smooth the data throughput transfer profile.