Clustering is essential to many tasks in pattern recognition and computervision. With the advent of deep learning, there is an increasing interest inlearning deep unsupervised representations for clustering analysis. Many workson this domain rely on variants of auto-encoders and use the encoder outputs asrepresentations/features for clustering. In this paper, we show that an l2normalization constraint on these representations during auto-encoder training,makes the representations more separable and compact in the Euclidean spaceafter training. This greatly improves the clustering accuracy when k-meansclustering is employed on the representations. We also propose a clusteringbased unsupervised anomaly detection method using l2 normalized deepauto-encoder representations. We show the effect of l2 normalization on anomalydetection accuracy. We further show that the proposed anomaly detection methodgreatly improves accuracy compared to previously proposed deep methods such asreconstruction error based anomaly detection.