Distributed generation (DG) has recently drawn the interest to meet the increased load demand with minimum investment. But cohesive operation of these DG sources, in a grid-connected environment, gives rise to several issues during abnormal conditions of the utility system. This paper addresses the detection method of one such crucial event which is “islanding”. A short length window based Mahalanobis Distance method has been proposed in this paper to detect islanding. A trade-off between computational time and accuracy has been maintained to make it reliable and acceptable. In this method, network parameters such as rate of change of frequency (ROCOF), rate of change of voltage (ROCOV), rate of change of real power (ROCOP) and rate of change of reactive power (ROCOQ) have been extracted from the voltage and current signal. Standard Deviations of the network features have been used as parameters for islanding and non-islanding events. These parameters have been classified with the proposed Mahalanobis Distance method incorporating short length window. The proposed method has been simulated in a test distribution system and it has been compared with Support Vector Machine (SVM), and Feed-forward Multi-layer Neural Network (FFML NN) classifiers to show its reliability and acceptability.