TCP congestion control algorithms have been design to improve Internet transmission performance and stability. In recent years the classic Tahoe/Reno/NewReno TCP congestion control, based on losses as congestion indicators, has been improved and many congestion control algorithms have been proposed. In this paper the performance of standard TCP NewReno algorithm is compared to the performance of TCP Vegas, which tries to avoid congestion by reducing the congestion window (CWND) size before packets are lost. The article uses fluid flow approximation to investigate the influence of the two above-mentioned TCP congestion control mechanisms on CWND evolution, packet loss probability, queue length and its variability. Obtained results show that TCP Vegas is a fair algorithm, however it has problems with the use of available bandwidth.