With the rapid development of the mobile Internet, social media dominated by Sina Weibo have become increasingly important in people's lives, thus the problem of credibility of social network information has come behind. As spread of rumors will bring a series of great harm, rumors detection is particularly important. Therefore this paper proposes a C-GRU model for rumors detection which combines convolutional neural network and gated recurrent unit networks. Our model firstly vectorizes the rumor events, then utilizes CNN to automatically construct the features of the rumor microblogs, and uses GRU to mine temporal information between associated microblogs under rumor events. The experimental results show that C-GRU model outperforms a series of classic models and is as well as superior to individual CNN or GRU.