The choice of stellar templates plays an important role in optical spectroscopic AGN classification, because the host galaxy contribution must be accurately subtracted in order to isolate the true contribution of the AGN. Up to now, simple stellar population models such as BC03, have been used as templates in doing the stellar component analysis. As more stellar population models become available, systematic study of the impact of the stellar population modeling becomes possible. This is important not only for finding the best template but also for understanding the merits and limitations of the templates. We analyzed the SDSS DR8 spectra, using different empirical, theoretical, and mixed stellar population models. We found that some templates lead to systematic biases in the identification of AGN candidates. We investigated the effects of the range of age,metallicity, and the total wavelength used in full-spectrum fitting. We found that the completeness of parameter space in the template model plays a vital role in classifying AGN candidates; the wavelength range used to analyze the spectra also affects the result but in a relative minor way. Empirical stellar models can be expected to yield the most reliable estimate of the absorption features in the host galaxies, since there will be less model dependence (e.g., on opacity assumption, line profile representation).