In this paper, a post-operational detection method based on functional principal component analysis and clustering is presented and compared with regard to designed operational criteria. The methodology computes an atypical scoring on a sliding window. It enables not only to detect but also to localize where trajectories deviate statistically from the others. The algorithm is applied to the total energy management, estimated from ground-based data, during approach and landing. The detected atypical flights show non-nominal energy behaviors such as glide interceptions from above or high speed approaches. This promising methodology could help to enhance flight data analysis and safety, highlighting non-monitored behaviors.