In job shop manufacturing environments, controlling the release of wor
k orders can significantly improve system performance, especially when
a few bottleneck resources limit shop throughput. A periodic order re
lease policy measures the remaining workload of the bottlenecks and re
leases just enough work orders to bring the workload to specified thre
shold values. The thresholds should limit the work-in-process inventor
y but not delay order completion by starving the bottlenecks. In this
paper we consider the problem of setting thresholds that minimize the
expected cost of work-in-process inventory and order tardiness. For th
e single-bottleneck system, we propose a stochastic optimization appro
ach that uses smoothed perturbation analysis to estimate the objective
function gradient. Our computational experiments compare different me
thods that set good threshold values and show that periodic order rele
ase improves system performance.