The normalized Laplacian spectrum is a good indicator of connectivity for comparing graphs with different sizes (i.e., the number of nodes). This paper shows the performances of several random-walk sampling algorithms using the spectral indicator. Based on two types of complex network models with exponential and power-law degree distributions, we determine that the sampling algorithms do not perform well on the spectral indicator, and find that the spectral indicator of biased sampling graphs is much closer to that of original graphs when the assortative mixing coefficient of original graphs decreases. Especially, for disassortative networks, biased sampling graphs perform better than unbiased sampling graphs on the spectral indicator.