The heart is one of the key organs of the human body. If its functions were disrupted for any reason, the quality of life is significantly reduced and the risk of death is increased. In order to prevent unwanted consequences, pathological changes in the ECG signal should be detected and the treatment should start as soon as possible. This can be achieved by detecting and analyzing characteristic parts of the ECG signal. QRS complex detection is used because the R peak is the most prominent part of each heartbeat so it is easily detectable. This paper describes the structure and activities of the heart. It also analyzes the electrocardiogram and how it is obtained. Noise removal using wavelet transform and QRS detection using phasor transform with sensitivity of 98,71% and positive predictivity of 99,09% are explained in detail. The paper also describes the k-nearest neighbors method used for classification of the detected QRS complexes. It distinguishes the premature ventricular contractions from the normal ones. After the training of the classifier on a balanced set of data, obtained sensitivity was 99,15%, and positive predictivity 96,82% with minor deviation depending on the randomly selected beats. After the training of the classifier on an unbalanced set of data, obtained sensitivity was 96,63% and positive predictivity 99,02%. These methods are implemented in MATLAB software package, while the testing is performed on the data obtained from the MIT-BIH Arrhythmia database.