The Chinese Han population, the largest population in the world, has traditionally been geographically divided into two parts, the Southern Han and Northern Han. In practice, however, these commonly used ethnic labels are both insufficient and inaccurate as descriptors of inferred genetic clustering, and can lead to the observation of “spurious association” as well as the concealment of real association. In this study, we attempted to address this problem by using 14 microsatellite markers to reconstruct the population genetic structure in 768 Han Chinese samples, including 384 Southern Han and 384 Northern Han, and in samples from Chinese minorities including 48 Yao and 48 BouYei subjects. Furthermore, with a dense set of markers around the region 5q34–35, we built fine-scale haplotype networks for each population/subpopulation and tested for association to schizophrenia susceptibility. We found that more variants in SLIT3 tend to associate with schizophrenia susceptibility in the genetically structured samples, compared to geographically structured samples and samples without identified population substructure. Our results imply that identifying the hidden genetic substructure adds power when detecting association, and suggest that SLIT3 or a nearby gene is associated with schizophrenia.