This paper is concerned with blocking probabilities in loss networks where both multicast and unicast connections are present. Each connection specifies a route and a bandwidth requirement. Multicast connections are point-to-multipont and allow transmission from one source to several receivers, any of which can join the connection at any time. Users that subscribe to the same data stream share one multicast connection so that no additional bandwidth is required. We present a Markov process model of a single network link. We also develop a convolution algorithm that finds the blocking probabilities on a single link and a reduced load approximation for estimating end-to-end blocking probabilities in a network. Blocking probability, loss network, multicast, reduced load approximation.