Breath sound recordings from pediatric subjects pose more processing complications. Children, especially the younger ones, are not able to follow instructions to stay calm during recording. This makes their recordings not only contain stationary artifacts but also non-stationary artifacts such as movement of subjects and their heartbeats. Further, the breath sounds from pediatric subjects also have lower magnitude compared to adults. In this work, we proposed to address those problems by developing a method to remove the artifacts from breath sound recordings. We implemented a combination of a Butterworth band pass filter and a discrete wavelet filter. We tested three types of wavelets (Coiflet, Symlet and Daubechies). Ten level decompositions and a set of hard thresholds were implemented in our work. Our results show that our developed method was capable of removing the artifacts significantly while maintaining the signal of interest. The highest signal to noise ratio improvement (10.65dB) was achieved by 32 orders Symlet.