Estimating spatio-temporal PM2.5 concentrations is a method that can be used to impute missing observations; which means that it may have important applications in the assessment of health effects. Based on observations gathered from 119 monitoring stations located within Henan Province and its surrounding areas, China, this study constructed a spatio-temporal dynamic model and used the Integrated Nested Laplace Approximation-Stochastic Partial Differential Equation (INLA-SPDE) method to estimate daily PM2.5 concentrations in Henan Province. Local geographical variables, such as elevation, road density, and land use, were integrated as spatial covariates. Meanwhile, meteorological variables, such as precipitation, air pressure, relative humidity, temperature, and wind speed, were integrated as spatio-temporal covariates. In addition, a first-order autoregressive process and a spatially correlated random effect were explicitly specified to capture the spatio-temporal dependence of PM2.5 concentrations. The validation results showed that the predictions were in good agreement with the observations (10-fold cross validated R2: 0.9407; Root mean square error: 10.7135 μg/m3), and the coverage probability of the 95% confidence interval was 96.04%. This study confirmed that the INLA-SPDE method can effectively model the spatio-temporal variability in PM2.5 concentrations. We also derived estimations of PM2.5 concentrations with a high spatio-temporal resolution, which should improve the assessment of related health effects.