A general model of Intelligent Communications Concentrating Devices (ICCD) is presented and analyzed for delay and overflow performance with compensation for the pipelining e f f e c t of message arrivals extending over time. The results of the analysis indicate that, for the same trunk utilization, the trend towards buffered terminals with longer messages requires substantially greater buffering in the ICCD. The nominal environment analyzed consisted of 10-40 medium speed terminals (1200 b/s -9600 b/s) operating over a medium speed trunk (9600 b/s) with trunk utilizations from 20 percent to 80 percent and average message lengths up to 1000 characters. This is a substantially different environment than that typically served by current implementations of ICCDs, which are frequently reported to have throughput improvements of 2-3 times the nominal originating terminal bandwidths, as opposed to the typical factor of 5 for the analyzed environment. This does not reflect on the appropriateness of the ICCDs in serving the new environment, but rather is simply stating that in the new environment the same character volume of traffic may be appearing with different traffic characteristics over higher speed access lines. If the new environment shows only a difference in traffic characteristics and originating line speed, without change in the traffic control scheme (or lack of scheme), the results indicate essentially reproduction of a large part of the terminal buffering in the ICCD for adequate overflow performance. Alternatively, with smarter terminals, traffic control schemes (flow control) may enable the ICCD to be reduced to an essentially unbuffered "traffic cop," with the terminal buffering also serving as the shared facility buffering. Several practical implementations of ICCDs have provision for flow control, but require cooperating terminals and hosts. This suggests that ICCD design and application will become more sensitive to the practical operating features of the target environment than has been generally the case to date. The analysis presented in this paper involves many simplifications to the actual problem. Additional work to accommodate non-exponential message length distributions and heterogeneous terminal configurations are perhaps two of the more immediate problems that may be effectively dealt with.1.