The rapid development of modern network technology, personal computers have gradually become popular, and private computer networks face more security threats. Among various network threats, malware threat is one of the main threats to personal computers. There are many types of malware, from simple worms and Trojan horse programs to complex computer viruses. The purpose and behavior of different types of malware are different. To better prevent the corresponding malicious behavior, We need to analyze malware features and classify them. In recent years, malware developers have continuously developed new technologies to evade traditional detection. This makes the number of malware families and malware variants continue to increase, and malware developers use polymorphism and metamorphism technologies extensively, making malicious executable files of the same family often modified or confused. Moreover, the static engine in the polymorphic malware can encrypt and decrypt the code, making it difficult for anti-malware technology to analyze the corresponding malware family through disassembly technology. So we need a simple and effective method to classify malware families. Visualization technology is an emerging technology proposed in recent years. Visualization technology allows us to observe the layout and texture of malware files directly. Observation shows that the visualized images of similar malware families have similar structures and textures. This paper proposes a novel, fast and effective method to visualize malware as RGB images and classify malware binary files into various malware families according to the visualized images. CNN is one of the algorithms often used in deep learning. And CNN has excellent performance in image processing. So we will use CNN as the classifier. We use the VGG-16 structure in the CNN model as our classifier. Experimental results show that our method achieves the highest accuracy of 98.9% in 9342 samples from 25 families. Compared with only using the grayscale image, the accuracy rate is increased by 2.5%, and the time cost is reduced by 9%-17%. Experiments have proved that our proposed method is effective and fast to realize the classification of malware families.