Abstract: A switching network with k=p+q inputs and as many outputs comprises one or more symmetrical time-division matrices, k being eight or a multiple of eight in the embodiments specifically disclosed. A larger number p of incoming subscriber links and a smaller number q of exchange outputs are respectively connected to p outgoing subscriber links and q exchange inputs. Digitized voice samples arriving over each incoming link in successive time slots of a recurrent frame, allocated to respective subscribers, are stored in each matrix and are selectively distributed to the reduced number of time slots sent to the exchange under the control of switching instructions from the exchange; conversely, signal samples received from the exchange in successive time slots are distributed to selected time slots of outgoing links. Signal paths can also be established between network inputs and outputs connected to incoming and outgoing links, thereby enabling direct communication between subscribers served by the network.