We propose to combine extension neural network type-3 (ENN-3) with the chaos theory and empirical mode decomposition (EMD) for electrocardiography (ECG) identification. ECG signals are measured and captured by the developed hardware measuring circuit and LabVIEW human–machine interface, and the stored ECG data are subjected to EMD at high- and low-frequency signals. A chaos dynamic error scatter map is formed using master and slave chaotic systems in order to obtain the chaos eye coordinates of a specific ECG signal, and ENN-3 is used for identification. There are 50 research participants in this study; the first half of the data are measured using a signal capturing circuit and a wrist patch-type ECG sensor (patch electrodes), while the second half are provided by Massachusetts Institute of Technology-Beth Israel Deaconess Medical Center (MIT-BIH). Analysis results show that the method proposed in this study has a higher accuracy in the classification of ECG signals, and that the recognition rate is as high as 100%. The recognition result was compared with those of ENN-3, the multilayer neural network, extension method, and ENN. The results showed that ENN-3 has a higher recognition accuracy rate than the other three algorithms, the difference being as much as 8%. Therefore, the autodiagnosis ECG system designed in this study can effectively classify arrhythmia and reduce the high cost of manual identification.