In the present work we study an statistical analysis of frequency spectrum as an anomaly detection technique. Our goal is to develop method that will able to detect malicious behavior in computer network. We are focused on detection of tunneled connections over HTTP protocol in order to circumvent restrictions by malicious agents.