Network performance anomalies can be defined as abnormal and significant variations in a network's traffic levels. Being able to detect anomalies is critical for both network operators and end users. However, the accurate detection without raising false alarms can become a challenging task when there is high variance in the traffic. To address this problem, we present in this paper a novel methodology for detecting performance anomalies based on contextual information. The proposed method is compared with the state of the art and is evaluated with high accuracy on both synthetic and real network traffic.