Convolutional neural network (CNN) is a kind of common architecture of deep learning, which is inspired by the biological visual cognition mechanism. CNN can obtain the effective feature expression from the original image. In recent years, CNN has made breakthroughs in the field of image recognition, but it takes a lot of time in the training process. As a new machine learning algorithm proposed by Leo Breiman in 2001, random forest (RF) has high accuracy in classification and regression, fast training speed and is not prone to over-fitting. The existing RF based classifiers rely on hand-selected features. Aiming at the above problems, a new C-RF model based on CNN is proposed in this paper, which puts the features extracted by CNN into RF to complete the classification.Since the network using random weights can also obtain effective results, gradient algorithm is not used to adjust the network parameters for avoiding a lot of time consumption. Experimental results on the MNIST and the Rotated MNIST datasets show that the classification accuracy of C-RF model is better than that of RF, and the generalization ability is also improved at the same time.