Implementations of classical statistical methods such as Holt-Winters filtering are widely available in statistical packages and open source software. Often these methods appear (at least visually) to model data effectively. However, the complexities of real time series data often render these methods ineffective for accurate anomaly detection and prediction. In this technical brief, we explore two approaches to time series data modeling and anomaly detection. We discuss how anomaly detection on real-world time series data requires specific modeling capabilities, and we compare anomaly detection results obtained with each approach.