Multiplying adjacent wavelet scales to obtain multiscale product can enhance significant structures and suppress noise, which makes it have predominance over other methods. Starting from the formation mechanism of multiscale product, soft-thresholding of multiscale product is proposed. Furthermore, we examine the mean square error (MSE) characteristics of soft-thresholding and hard-thresholding based on multiscale product coefficient. Finally, we present a measurable rule to choose wavelet bases. Simulation results are consistent with the theoretic analyses.