The paper presents a new approach for estimating the throughput of a c
losed queueing network with exponential servers, finite buffer capacit
y and a blocking after service discipline. The problem is tackled by d
ecomposing the network. The population constraint is enforced by requi
ring that the sum of the expected number of customers in the various s
ubsystems is equal to the population size. Each subsystem is analyzed
as an M/M/1/C-i + 1 queue with state-dependent arrival and service rat
es. The rationale behind this last assumption is that the behavior of
the system at a given time is affected by the history of blockings and
starvations. The results obtained by applying the proposed algorithm
to a set of test problems show a good agreement with those obtained wi
th simulation, the difference on the maximum throughput of the network
being of the order of 3%. The obtained results also compare favorably
with those described in the literature.