Abstarct: ECG Signal compression is one of the interesting fields in telemedicine to improve recognition systems, it also reduces memory and bandwidth requirements which is need to store diagnostic information. ECG compression without removing diagnostic information is baxsed on the fact that digitized sequential samples of ECG signal carry additional information that can be removed by computational and statistical methods. In this thesis we present an ECG compression algorithm baxsed on pattern matching technique. The algorithm is tested on MIT_BIH arrhythmia databaxse and we evaluated the proposed technique by computing CR and PRD parameters on the selected records. Initially we need to compute the exact location of QRS complexes, so we applied Pan_Tompkins QRS detection method (which has the most accuracy compare to others) on the records, the result of this term is a vector containing QRS locations information. According to the proposed method, pattern matching is the second step, the baxse of this technique is descxription of original signal according to some patterns due to reduce redundant information and keep diagnostic information. In this part of algorithm we use DTW method to measure differences between patterns and ECG cycles. We need to consider a threshold for each record that is done manually, if the difference is less than the threshold, the ECG cycle replace by the pattern, otherwise the ECG cycle add to the library as a new pattern; the library forms in this way. The outputs of this term are considering as library prototypes, residual signal (the difference between ECG cycles and the patterns), and the pointers. The library prototype signals is encoded using 2D image coding using SPIHT coding, the residual signal coding is baxsed on DCT transform and arithmetic coding, and the pointers are encoded using arithmetic coding method, the coding step compress the signals that we reconstruct the ECG signal using them, and compute CR and PRD of MIT_BIH records after reconstruction. The best PRD is obtained 0.3025 for MIT_BIH record 100, and the best compression ratio is 3.2374 that belongs to record 221.We also compared the R_R intervals after and before compression, and record 103 had the least difference.