This article presents the problem of classification of endogenous rhythms with electrocardiography signals coming from patients with implanted cardiac pacemaker. Efficiency of detection of QRS complex was examined by algorithms working in time domain. During the investigation attention was paid to proper selection of level decomposition, good choice of detection threshold as well as choice of wavelet transformation. In case of identification of endogenic rhythm attention was paid to architecture of feedforward neural network, selection of teaching file and the accuracy of classification depending on the activation function used.