Honeypots are decoys designed to trap, delay, and gather information about attackers. We can use honeypot logs to analyze attackers? behaviors and design new defenses. A virtual honeypot can emulate multiple honeypots on one physical machine and provide great flexibility in representing one or more networks of machines. But when attackers recognize a honeypot, it becomes useless. In this paper, we address issues related to detecting and "camouflaging" virtual honeypots, in particular Honeyd, which can emulate any size of network on physical machines. We find that an attacker may remotely fingerprint Honeyd by measuring the latency of the network links emulated by Honeyd. We analyze the threat from this fingerprint attack based on the Neyman-Pearson decision theory and find that this class of attack can achieve a high detection rate and low false alarm rate. In order to counter this fingerprint attack, we make virtual honeypots behave like their surrounding networks and blend in with their surroundings. We design a camouflaged Honeyd by revising a small part of the Honeyd toolkit code and by appropriately patching the operating system. Our experiments demonstrate the effectiveness of our approach to camouflaging Honeyd.