The electrocardiogram (ECG) signals represent the electrical activities of the heart. Noise and artifacts are inherent contaminating components of the ECG signal and denoising of these signals is quite an important issue, since the noise can hinder accurate diagnosis and detection of cardiac diseases. Also, automated noise detection and classification and selective removal of noises can play a vital role in the development of robust unsupervised ECG analysis systems. This paper mainly deals with the comparison of two different approaches for noise detection and classification of ECG signals. A unified framework is followed for noise detection and classification which consists of a signal decomposition method, short term temporal feature extraction and decision rules for detecting the noises. The two approaches for noise detection and classification only differ at the decomposition step, one uses the complete ensemble empirical mode decomposition and the other uses the wavelet packet decomposition. The decomposed signals are then analysed using certain short term temporal features for automatic detection, localization and classification of single and combined ECG noises. This framework is evaluated on MIT-BIH arrhythmia database and results for both the cases are compared. Results show that, the performance of wavelet based decomposition is comparable to CEEMD. This enables the proposed wavelet decomposition based approach to be an ideal candidate for real-time continuous monitoring systems such as ECG Holter monitors. © 2019 IEEE.