In the present work, cross Hyperbolic S-transform (XHST) is proposed as a novel feature extraction technique from EEG signals for automated detection of epilepsy. XHST is proposed as an extension of hyperbolic Stockwell transform (HST) which is an effective tool to analyze any non-stationary signal in joint time frequency frame. In the present work, EEG signals of healthy (H) and seizure (S) subjects were taken from a publicly available online database and XHST of the respective signals were done separately with a reference H signal. Several meaningful features were extracted from the respective cross spectrums, from which only two significant features were selected after Kruskal-Wallis test. The selected features were henceforth given as inputs to a k-nearest neighbor (kNN) classifier for the classification of EEG signals. Investigations revealed that the proposed XHST based feature extraction technique delivered reasonably accurate results in detecting H and S EEG signals, which can be potentially implemented in the field of neuroscience for automated epilepsy detection.