A QRS detection algorithm has been developed based on a multivariate model. It included three normalized measures - amplitude, first difference, and spatial frequency. The terms are combined in a weighted sum to generate an indicator variable that is then compared to a detection threshold. An FIR, band-pass filter, consisting of a cascaded series of running medians and means prefiltered the raw ECG signal to reduce high frequency noise as well as low frequency base line drift. The technique appears to be exceedingly robust, correctly detecting even aberrant QRS complexes in noise-corrupted ECGs.