This thesis discusses the development of methods for computer analysis of cardiac arrhythmias. The study, which was undertaken at Glasgow Royal Infirmary, was designed to supplement the existing computer system for routine interpretation of orthogonal lead electrocardiograms (ECGs). A brief historical review of electrocardiography, including recent developments in the field of automated EGG interpretation, is initially presented. Thereafter, details of the methods used routinely in Glasgow Royal Infirmary at the commencement of the study are given together with a description of the computer facilities and allied equipment which were available at the outset of the investigation. These existing facilities, in conjunction with daily routine batch processing of ECGs for interpretation, provided material for continuous evaluation of the developing arrhythmia analysis program. The commonly encountered arrhythmias are outlined together with an account of the diagnostic criteria used by the program for their classification. Although not all complex and compound arrhythmias could be accommodated in the logic, the rhythm statements of the program cover the large majority of ECGs encountered in a routine hospital service. The author agrees with other workers that the more urgent need at present is to reduce the incidence of erroneous interpretations among the common, simple arrhythmias. The important wave-recognition procedures and associated decision tree are documented in the form of flow diagrams. All programming was done in assembly language, which, it is argued, should be used for this particular application, ECG data are preprocessed by digital filtering. There being several classes of digital filter, and in general many algorithms for realising a desired frequency response, the merits of each method had to be assessed in turn. From the dedicated nature of the ECG system it followed that execution time and memory storage requirements were significant factors in finally selecting a high speed recursive filter with an integer-arithmetic implementation for use in the arrhythmia program. In each of three surveys of the program's accuracy, human and computer interpretations of all ECGs processed during the sampling period were stored. Results for approximately 5000 cases were filed on digital magnetic tape and subsequently analysed by FORTRAN programs. Figures of program accuracy for individual and combined arrhythmias are discussed, as are the various sources of error. In order to draw some comparisons, similar figures are quoted for arrhytlimia programs developed elsewhere. Reliability of negative reporting was emphasised in the formulation of diagnostic criteria, whence errors of under-interpretation are much rarer than those of overinterpretation. The latter can easily be discounted by the physician, whereas the former might be overlooked. Statistics of program performance are not maximised by such a failsafe policy. Nevertheless the Arrhythmia Program achieved an accuracy rate of 95% for all ECGs analysed, and 85% correct interpretation for the arrhythmia subgroup. The thesis therefore contends that arrhythmia diagnosis on a small dedicated digital computer can enhance the value of routine ECG reports produced by an automated interpretation system.