To explore the evaluation of white matter structural network analysis in the differentiation of Alzheimer's disease (AD) and subcortical ischemic vascular dementia (SIVD), 67 participants [31 AD patients, 19 SIVD patients, and 19 normal control (NC)] were enrolled in this study. Each participant underwent 3.0T MRI scanning. Diffusion tensor imaging (DTI) data were analyzed by graph theory (GRETNA toolbox). Statistical analyses of global parameters [gamma, sigma, lambda, global shortest path length (Lp), global efficiency (Eg), and local efficiency (Eloc)] and nodal parameters [betweenness centrality (BC)] were obtained. Network-based statistical analysis (NBS) was employed to analyze the group differences of structural connections. The diagnosis efficiency of nodal BC in identifying different types of dementia was assessed by receiver operating characteristic (ROC) analysis. There were no significant differences of gender and years of education among the groups. There were no significant differences of sigma and gamma in AD vs. NC and SIVD vs. NC, whereas the Eg values of AD and SIVD were statistically decreased, and the lambda values were increased. The BC of the frontal cortex, left superior parietal gyrus, and left precuneus in AD patients were obviously reduced, while the BC of the prefrontal and subcortical regions were decreased in SIVD patients, compared with NC. SIVD patients had decreased structural connections in the frontal, prefrontal, and subcortical regions, while AD patients had decreased structural connections in the temporal and occipital regions and increased structural connections in the frontal and prefrontal regions. The highest area under curve (AUC) of BC was 0.946 in the right putamen for AD vs. SIVD. White matter structural network analysis may be a potential and promising method, and the topological changes of the network, especially the BC change in the right putamen, were valuable in differentiating AD and SIVD patients.