This paper outlines a higher-order statistics (HOS)-based technique for detecting abnormal conditions in voltage signals. The main advantage introduced by the proposed technique refers to its capability to detect voltage disturbances and their start and end points in a frame whose length corresponds to, at least, N = 16 samples or 1 / 16 of the fundamental component if a sampling rate equal to f s = 256 × 60 Hz is considered. This feature allows the detection of disturbances in submultiples or multiples of one-cycle fundamental component if an appropriate sampling rate is considered. From the computational results, one can note that almost all abnormal and normal conditions are correctly detected if N = s256, 128, 64, 32, and 16 and the SNR is higher than 25 dB. In addition, the proposed technique is compared to a root mean square (rms)-based technique, which was recently developed to detect the presence of some voltage events as well as their sources in a frame whose length ranges from 1 / 8 up to one-cycle fundamental component. The numerical results reveal that the proposed technique shows an improved performance when applied not only to synthetic data, but also to real one.