This paper deals with traffic signal control with finite queue capacities in a discrete-time and stochastic setting. A so-called “pressure releasing policy” (PRP) is introduced to optimally release traffic pressure at every time slot, where the traffic pressure at each intersection incorporates knowledge of turning ratios and information of neighboring and ingress queues. PRP does not require knowledge of arrival rates. Moreover, it employs a set of weights satisfying a given condition to handle downstream queue spillover, and an algorithm is provided to generate one possible set of weights. Define the throughput region as the closure of the set of all arrival rate vectors that can be stably supported over the network under the assumption on infinite queue capacities. It is shown that PRP under finite queue capacities can still achieve the closed-loop stability with a reduction on the throughput region. The reduction is a function of weights and internal queue capacities, and PRP with finite but sufficiently large internal queue capacities can be arbitrarily close to recovering the throughput region.