In this paper, we study the problem of determining the capacity requirements for applications that require QoS guarantees. We consider three kinds of source models; Poisson based, On-Off based and leaky bucket filter based, and present explicit expressions capturing the required capacity. We compare the required capacity for average delay for these models with the required capacity for absolute delay and comment on the differential. Further, using simulation results, we study the average and variance of the observed delay for voice and video sources and compare the three models. We found that On-Off and leaky bucket models are very powerful and ensure that the actual delays are less than the required delay and that the variance remains acceptable. These results seem to indicate that it may be possible to dimension systems based on average QoS requirements and still get adequate performance for other requirements such as jitter. This would then provide us with computational tools to dimension networks efficiently.