Connection Admission Control (CAC) algorithms are used to decide whether an incoming connection should be accepted or rejected in a node of a network offering reservation based services in order to maintain the guaranteed Quality of Service (QoS) in the network. In this paper, we consider the statistical CAC algorithm proposed by Elwalid et al. [ 2 ]. The traffic model is made of ON-OFF sources and the QoS parameter is the loss probability. Based on the traffic descriptors of existing and incoming connections, the algorithm takes its decision by computing an upper bound of this probability and checking whether it is larger than a given tolerance . Usually this tolerance is a fixed, given parameter. We propose here to adapt to react to the actual losses experienced at the node using a simple regulation mechanism: if the actual loss ratio is much smaller than the targeted loss ratio, is increased to make a more aggressive usage of the available resources, and vice versa if the actual loss ratio is too high. We discuss the influence of the regulation parameters and we show that despite its simplicity this regulated CAC improves significantly the performance of its non-tunable counterpart.