We propose a new compression technique which exploits the high correlations between the consecutive beats of an Electrocardiogram (ECG). A single progressive average beat (PAB) and residues were computed from every 10 ECG beats and subjected to compression using Discrete Wavelet Transform (DWT). Reconstruction errors were controlled by selective reconstruction of the wavelet coefficients from the PAB and residues using 5% PRDN and 99% energy reconstruction efficiency (ERE) criteria. Finally, the selected coefficients were encoded using a lossless compressor, a combination of delta and run length encoder. An average CR and absolute maximum error of 8.96, and 0.08 mV respectively were obtained with 32 numbers of MITA data under Physionet. Major clinical features like QRS amplitude, T height, P height and QT interval had low distortion when the target PRDN limit was relaxed up to 15%. © 2018 IEEE.