Cyber criminals congregate and operate in crowded online underground marketplaces. Because forensic investigators lack efficient and reliable tools, they are forced to analyze the marketplace channels manually to locate criminals – a complex, time-consuming and expensive task. This paper demonstrates how machine learning algorithms can be used to automatically determine if a communication channel is used as an underground marketplace. Experimental results demonstrate that the classification system, which uses features related to the cyber crime domain, correctly classifies 51.3 million messages. The automation can significantly reduce the manual effort and the costs involved in investigating online underground marketplaces.