In this work, we propose a new sliding mode controller based on a reference model for controlling data transmission rates in a connection-oriented communication network. In the proposed approach, we build a model of the network, which is controlled by a linear quadratic (LQ) optimal controller. Then, the sliding variable of the real network is forced to follow the reference sliding variable generated by this model. This method enables us to preserve the favorable properties of the optimal controller, such as reducing and smoothing out the initial flow rates. Moreover, once the reference model attains the vicinity of the desired state, the designed sliding mode controller is able to react more rapidly to changes in the available bandwidth, ensuring better robustness. Due to this fact, the considered controller can ensure full bottleneck link bandwidth utilization with smaller memory buffers than the LQ optimal one. This important advantage is shown both analytically, and in computer simulations K E Y W O R D S communication systems, model reference control, sliding mode control 1 INTRODUCTION Sliding mode control is a well-developed control approach that has been quite thoroughly studied over the years. The majority of publications deal with continuous time systems, 1-8 and in some of them, model reference sliding mode schemes are proposed for such systems. 9-13 However, nowadays even if the plant itself is described in continuous time, it is usually controlled by a digital device (eg, programmable logic controller or a microcontroller). Therefore, to fully take into account the impact of discretization of the control signal, the discrete time sliding mode control has been proposed. 14-22 Unfortunately, discrete time model reference control has not been studied in detail. Therefore, in this work a model reference based discrete time sliding mode controller is proposed. The controller will be used to determine transmission rates in a connection-oriented communication network, 23-27 such as networks operating under MultiProtocol Label Switching standard. It is assumed, that during data transmission from multiple sources to multiple destinations a single bottleneck occurs, due to an insufficient transmission capability of a single network link. In such a case, the data is stored at the node immediately before this link, which will be called the bottleneck node. This node must then send information to the data sources, to limit their transmission rates and prevent data loss due to buffer overflows. The proposed methodology begins with building a mathematical model of the network. The next step is designing an optimal controller 28,29 which will control this model. The sliding variable evolution obtained from the model will then be