Internet represents a shared resource, wherein users contend for the finite network bandwidth. Contention among independent user demands can result in congestion, which, in turn, leads to a long queuing delay, packet losses or both. In the Internet, there are two mechanisms which deal with congestion, the end-to-end mechanism which is achieved by the Transmission Control Protocol (TCP) and the intermediate nodes algorithms such as Active Queue Management (AQM) in routers. In this thesis, a combined model of TCP and AQM (TCP/AQM) is presented and first simulated without a controller. The results show that it is unable to track the desired queue size. To get a better tracking performance the Proportional Integral (PI) controller was used as AQM in the router queue, which shows a good tracking performance but with an overshoot 28.66% and a settling time of 28 sec. To reduce the overshoot and to speed up the system response, first a Fuzzy Neural like PI controller is designed for TCP/AQM router. The parameters of Fuzzy Neural like PI controller are selected at first by trial and error process method and then by using the optimization techniques (Genetic Algorithm (GA) or Particle Swarm Optimization (PSO) algorithm) in order to get the best controller parameters and to enhance the overall system response. The PSO method was better than GA which gives 1.66% overshoot compared with 5% overshoot using GA. A Fuzzy Neural like PID controller was then designed for TCP/AQM router. The parameter of Fuzzy Neural like PID controller were also selected first by trial and error process method and then by using optimization techniques (GA or PSO) and it also showed that PSO method was better than GA which gives zero overshoot compared with 0.6% using GA and 7 sec settling time compared with 15 sec using GA. The simulation results for TCP/AQM model were presented in MATLAB version 9.0. The controller simulation result shows the ability of the Fuzzy Neural like PID controller to track the desired queue size with zero overshoot and shows that the use of PSO reduces the computational time compared with GA by 87.2%.