We implemented a method for compression of the abulatory ECG that includes average beat subtraction and first differencing of residual data. Our previous investigations indicated that this method is superior to other compression methods with respect to data rate as a function mean-squared-error distortion. Based on previous results we selected a sample rate of 100 samples per second and a quantization step size of 35 microV. These selections allow storage of 24 h of two-channel ECG data in 4 Mbytes of memory with a minimum rms distortion. For this sample rate and quantization level, we show that estimation of beat location and quantizer location can significantly affect compression performance. Improved compression resulted when beats were located with a temporal resolution of 5 ms and coarse quantization was performed in the compression loop. For the 24-h MIT/BIH arrhythmia database our compression algorithm coded a single-channel of ECG data with an average data rate of 174 bits per second.