Several studies concerning the performances of TCP over UBR service have shown that ATM switches with limited buffers respond to UBR congestion with a low throughput and high unfairness. In order to achieve a higher efficiency it is possible to implement some additional mechanism to control cells dropping at buffers switches. One of the most popular drop strategy is the early packet discard, which drops the entire higher level data units when the buffer queue reaches a certain threshold level. Another mechanism is the partial packet discard, which, if a cell is lost in case of congestion, discards all the subsequent cells belonging to the same packet. The first investigations of the throughput behavior of TCP over ATM with EPD, using simulation studies, have shown a performance improvement. How efficiently the switches buffers are used depends on the placement of the EPD threshold and on how cell dropping will occur, due to the level of the congestion. Two of the most relevant factors are the distribution of the packet size and the traffic distribution. Our main goal is to analyze a relation between the efficiency for different buffer dimensions and the parameters, as the packet size and the number of sources on the network, responsible for the volume of the traffic. Furthermore we want to investigate the optimal EPD threshold value. The performances with and without application of the enhancements are compared, trying to supply a relation by interpreting the results obtained with the simulations. This suggests an answer to the question 'why and when early packet discard' could be used.