Abnormal variations of traffic are conventionally considered to occur under the condition that traffic rate is abnormally high in the cases, such as traffic congestions or traffic under distributed denial-of-service (DDOS) flood attacks. Various methods in detecting traffic variations at abnormally high rate have been reported. We note that a recent paper by Kuzmanovic and Knightly, which explains a type of DDOS attacks that may result in abnormally low traffic rate. Such a type of abnormal variations of traffic, therefore, can easily evade from detection systems based on abnormally high traffic rate. This paper presents a real-time and reliable detection approach to detect traffic variations at both abnormally high and low rates. The formulas in terms of detection probabilities, miss probabilities, classification criterion, and detection thre-sholds are proposed.