Traditional seizure and interictal detection techniques aim to quantify the amount of activity present. To do this they must have a high sensitivity (to correctly detect all of the events) and few false detections (high specificity). These requirements are difficult to fulfil simultaneously. An alternative approach, termed data selection, is presented here. This method is a different view on the detection problem: A detection procedure is used to select which sections of data are saved—a fixed amount on either side of an automated detection—and which are discarded. A high sensitivity is still required, but false detections are not as significant as they are rejected by a human interpreter in the same way as background signals are when a standard continuous EEG trace is analysed. The method only allows data reduction, not event quantification, but this is still significant in reducing the analysis time required and in facilitating wireless ambulatory EEG units. Ordinarily there is too much raw EEG data to transmit without compromising battery life and so online, low power, data reduction is required. The data selection algorithm’s tolerance to false detections simplifies the algorithm design making it very suitable for this low power implementation. The data selection method presented here is based upon the Continuous Wavelet Transform and can offer a 50% reduction in the amount of data to be transmitted whilst correctly recording 95% of expert marked interictal events. All of the algorithm blocks are also suitable for ultra low power VLSI implementation.