Background: Automatic detection and classification of cardiac abnormalities in the ECG is one of the basic and often solved problems. The aim of this paper is to present a proposed algorithm for ECG classification into nine classes Normal, AF, I-AVB, PAC, PVC, LBBB, RBBB, STD, and STE. Method: Signals from the PhysioNet / CinC Challenge 2020 database were used. Detector of QRS complexes and P-waves including detection of their beginnings and ends was designed. The detector combines phasor, wavelet and s-transform. Then, the most common morphology of the QRS was found in each record. All of these QRS were averaged. Pathologies are detected according to the following rules: LBBB: the morphology of the QRS is rS or QS in leads V1 and V2 and QRS duration > 120 ms; RBBB: QRS is predominantly positive in lead V1; I-AVB: PR interval > 196 ms and the signal is not AF; PVC: correlation of QRS and pattern and analysis of RR interval length; PAC: analysis of RR intervals and the signal is not AF; AF: bagged tree classifier; STD and STE: SVM classifier. Both created classifiers were validated on a training dataset using a 10-fold cross-validation. Signal was classified as Normal if no pathology was found. Results: The proposed algorithm has the following results on the training set: F-2: 0.9565, 0.8388, 0.8441, 0.8611, 0.8025, 0.8064, 0.8800, 0.8248, 0.6594 and G-2: 0.8774, 0.5969, 0.6020, 0.6529, 0.5103, 0.5282, 0.7035, 0.6152, 0.4164 for AF, I-AVB, LBBB, Normal, PAC, PVC, RBBB, STD and STE, respectively. The final F-2 and G-2 on the training set is 0.830 and 0.611. Classification results on the hidden test database is F-2 0.785 and G-2 0.558. Conclusion: The presented method classifies pathologies using low number of significant features designed for each pathology.