We propose a method to quantify the statistical significance of anomalies detected by deep k-nearest neighbor approach.