The aim of this study was to develop methods and systems which could improve the accuracy or arrhythmia monitoring and analysis of long-term ECGs and eliminate or reduce the need for continuous visual ECG screening. The detection of arrhythmias was based on an automatic classification of heart beats, using waveform features obtained from a basis signals representation. Ventricular fibrillation, a condition in which individual QRS complexes cannot be discerned, was diagnosed from a spectral analysis of the ECG.The methods were put into practice in a computer-based system with the capacity for simultaneous monitoring of eight patients. The accuracy of the system with respect to arrhythmia alarms was studied during ten 24-h periods of clinical routine use, with 55 patients. monitored for a total time of about 1000 h. Seventy percent of the time during which an alarm message should have been present (134 h) a correct alarm was given. An incorrect alarm cause was reported 17% of the time, while 13% of the time no alarm at all was given. The ratio between true and false or incorrect alarms was 3:1.The methods for arrhythmia detection were also utilized in a computer program for offline analysis of long-term ECGs from ambulatory patients. Here, the automatic analysis was combined with a subsequent interactive examination of the analysis results for highest accuracy.