In the previous work, we have designed and implemented a platform with tools for capturing malware, running botnets in a controlled environment, analyzing their interactions with a botmaster, testing methods and techniques for mitigating botnet nuisance, and eventually disrupting them. We have used the platform to gather a large number of malware and observe its network activity. In this paper, we present an approach to malware classification based on the observation of the malware communication behavior. First, we show that traditional methods based on antivirus tools are not suitable for classification. Then, we define the method based on observing the communication pattern of executing malware. We report on the classification results obtained with the proposed method. Unlike classification done by existing antivirus tools, the proposed method results in selective and consistent classification.