Current large-scale structure surveys mostly focus on the overdesne regions of the universe. However, the dominant regions in the large-scale structure are of low density compared to the mean value and the clustering of the underdense regions may have information complementary to that of the halo clustering. Comsic voids are large underdense regions in the large-scale structure. In this this talk I will discuss the clustering of voids, especially their linear bias, quadratic bias, and the void bias in the primordial non-Gaussianity scenario. In the second part of the talk, I introduce the volume statistics as a probe of the distribution of the underdense regions. Compared to voids, this statistic is less sensitive to the shot noise contaminations and could be easier to model. Using simulations, the bias of the volume statistics is meansured and determined to be negative. The Baryon Acoustic Oscillations are also measured and there is no systematic shift in the position.