Abstract: This paper explores the characteristics of micro pores at multi-scale using the scanning electron microscope(SEM) and determines the best magnification for quantitative analysis of pores. It analyses the SEM images of remolded loess samples with eight dry densities at 12 different magnifications(100X, 200X, 300X, 400X, 500X, 600X, 800X, 1000X, 1200X, 1500X, 2000X, 3000X)with IPP 6.0 software. It statistically classifies the loess pores according to Lei Xiangyi's classification standard. The results show the follows. When the magnification is less than 400X,the micropores can not be observed in any SEM image, and the macropores can not be observed at the magnification greater than 1000X. However, four different types of pores can be basically observed when the magnification is between 400X and 1000X. Based on the cluster analysis, the classification results of magnification are consistent with the statistical results of SEM images processed with IPP 6.0 software. The difference in multi-scale pore characteristics of samples with eight dry densities is mainly due to the transformation of different pore types with the change of soil compactness. With the increase of magnification, the field of view becomes narrower and the macropores are more difficult to be counted. On the contrary, the smaller the magnification is,the larger the visual field is, and the more easily the micropores are to be ignored. Considering the distribution and content of micropores and macropores, the best observation multiple is between 500X and 800X. This study not only helps to reveal the pore characteristics of remolded loess at multi-scale based on SEM,but also has reference value for quantitative characterization of soil microstructure in the field of geological engineering.