In most coherent imaging modality, speckle noise is a major cause that blurs the boundary of tissues and degrades the image contrast. By utilizing the unique properties of supercontinuum (SC) generated by noise-like pulses (NLPs) and a simple multi-frame averaging technique, we achieved significant speckle reduction in spectral domain optical coherence tomography (SD-OCT). We quantitatively compared the speckle of our proposed method with those of conventional swept source OCT (SS-OCT) and SD-OCT based on commercial light sources. The experimental results show that SC pumped by NLPs combined with noncoherent averaging method achieves better denoising performance in terms of contrast to noise ratio (CNR).