An electrocardiogram (ECG) is measured from the body surface and is often corrupted by various noises, such as high-frequency muscle contraction. Recently, empirical mode decomposition (EMD), a well-known analysis technique for nonlinear and non-stationary signals, has been employed for the purpose of ECG noise reduction. In this study, a modified EMD, ensemble empirical mode decomposition (EEMD), was used for ECG noise reduction. Additional Gaussian noise was applied, followed by the EMD process, and the average (ensemble) intrinsic mode function (IMF) was used for ECG reconstruction. In this study, three high frequency ECG noises, muscle contraction, 50/60 Hz power line interferences and simulated Gaussian noise were examined. Mean square error (MSE) between filtered ECG and clean ECG was used as a reconstruction performance index. Results showed that the first or the first two IMF levels were deleted owing to their noise components, whereas the other ensemble IMF constituted clean ECG components for ECG reconstruction. The MSE of EEMD is lower than the MSE of EMD and infinite impulse response (IIR) filter on these three noise types due to the reduction of mode-mixing effect between separate IMFs. Thus, the proposed EEMD-derived noise reduction performance was observed to be superior to the traditional EMD and IIR filter approaches.