The　non-stationary　characteristics　ECG　and　its　information　criterion　for　diagnosis　are　studied.　The　data　representation　model　and　the　compression　algorithm　are　studied　by　time-series　modeling　theory.　Using　wavelet　networks　as　data　representation　model,　a　compression　algorithm　of　ECG　is　proposed.　While　the　original　data　of　ECG　are　framedin　real-time,　the　algorithm　maps　the　data　of　each　frame　to　parameters　of　wavelet　networks　as　rebuilding　information　from　original　data.　The　principle　of　wavelet　networks　as　data　representation　model　and　the　compression　algorithm　based　on　data　framing　are　particularized,　then　the　experiment　on　the　compression/recovery　of　ECG　data　using　the　propesed　method　is　performed.