E-mailing as one of the most popular Internet communication service is very prone to spamming, i.e. the process of flooding our mailboxes with unsolicited messages. Every day, separating good messages from those coming from spammers we waste time and resources. It is urgent to find out an effective method of fighting with spam that should take into account not only the currently used spammers? methods but also dynamic aspects of constantly changing spammers? behavior.

The paper presents an on-line e-mails classification method that assesses every incoming e-mail as spam, non-spam or leaves it unclassified. The method uses various types of users? email communication graphs built from mail servers log files. Additionally the method uses comparative procedures that let us take into account dynamic aspects of spam process by the analysis of the predefined graphs built for consecutive periods of time. It is also presented how this method can and should be combined with today most efficient spam fighting methods using heavy content-based analysis tools. It is pointed out why these two methods should be considered as combined ones.