Spam Over Internet Telephone can become a serious threat within the close to the future due to the growing number of VOIP users. We tend to propose a model that uses the time duration (length) of calls between two users to differentiate normal callers and spam callers. Here, we define a process to notice spammers in VoIP by classifying differences in the Call Graph. This directed, by using call data records weighted graph and it is created where the set of separating call parameters are used to derive weights on the edges This system provides us a straight forward and a simple way to use call period as associate automatically selected model based on human behaviour.