This paper presents an efficient dissemination algorithm suitable for scale-free random topologies which model some complex real world networks. In these topologies, some sites, denoted hubs, have many more connections than the others. By exploiting then the dissemination power of hubs, we propose a new gossip algorithm where sites directly connected to hubs do not forward received messages. Our algorithm offers a very high reliability and does not require any input parameter value that informs each site if it is a hub or not. Such an information is deduced by every site during the algorithm execution. Compared to well-known probabilistic gossip algorithms, performance simulation results show that our algorithm presents good performance in terms of message complexity and latency.