The presence of a fragmented QRS complex (fQRS) in two contiguous leads of a standard 12-lead electrocardiogram (ECG) has been shown to be an indicator of myocardial scar in multiple different populations of cardiac patients. QRS fragmentation is also a predictor of adverse prognosis in acute myocardial infarction, coronary artery disease, and ischemic cardiomyopathy and a prognostic tool in structural heart diseases. An increased risk of sudden cardiac death associated with fQRS has been documented in patients with ischemic cardiomyopathy and hypertrophic cardiomyopathy. However, fQRS is also frequently observed in apparently healthy subjects. Thus, a more detailed classification of different QRS fragmentations is needed to identify the pathological fragmentation patterns and refine the role of fQRS as a risk marker of adverse cardiac events and sudden cardiac death. In most studies fQRS has been defined by the presence of an additional R wave (R′), or notching in the nadir of the S wave, or the presence of >1 R′ in two contiguous leads corresponding to a major coronary territory. However, this approach does not discriminate between minor and major fragmentations and the location of the fQRS is also neglected. In addition to this, the method is susceptible to large interobserver variability. We suppose that some fQRS subtypes result from conduction delays in the His-Purkinje system, which is a benign finding and thus can weaken the prognostic values of fQRS. The classification of fQRSs to subtypes with unambiguous definitions is needed to overcome the interobserver variability related issues and to separate fQRSs caused by myocardial scarring from benign normal variants. In this paper, we review the anatomic correlates of fQRS and the current knowledge of prognostic significance of fQRS. We also propose a detailed fQRS classification for research purposes which can later be simplified after the truly pathological morphologies have been identified. The research material of our study consist of 15,245 ECGs from the random general population and approximately six thousands (n = 6,241) ECGs from subjects with a known cardiac disease.