Development of noise cancellation algorithm is essential to facilitate accurate detection of electrocardiogram (ECG) in mobile health and wearable medical devices. In this study, we captured ECG from 20 subjects when they were at rest and during routine activities. The motion artifact in ECG was filtered using two non-linear filters: median filter and discrete dyadic wavelet transform. Signal-to-noise ratio (SNR) and computation time of the filters were determined. We found that median filter showed larger SNR (7.61±1.21 dB) than discrete dyadic wavelet transform did (5.35±1.34 dB). Conversely, discrete dyadic wavelet transform benefited to its short computation time. The algorithms of these non-linear filters should be further investigated to achieve both high SNR and fast computation in wearable and mobile monitoring systems.