A technique for detecting SPAM calls in VoIP networks and taking responsive actions to such calls is described. Detection is based on recognizing abnormalities in signaling message statistics. For example, a device that persistently sends many more call setup requests then it receives and at the same time constantly receives too many, or too few, call terminations in a statistically determined period of time is assumed to be a spam source. This criteria is based on the principle that a Spammer originates many calls, which are consistently terminated by the Spammer or by the recipient. Responsive actions to the detection of VoIP spam may include displaying a call text warning to the recipient, using special ring tone, automated voice mail answer, rejecting the call, etc. The invention addresses an issue that will become more problematic for VoIP networks and their subscribers in the future, and to which a solution has not been proposed in the prior art.