Data assimilation (DA) provides a way for effective combination of model simulation and observation, and improves accuracy of winter wheat yield estimation. Among various DA methods, the particle filter (PF) is not constrained by the conditions of linear models and Gaussian error distribution, and receives more attention and application of DA. Currently, most researchers adopt single remotely sensed data source and single variable assimilation strategy, which cannot accurately reflect the interactive process among radiation, temperature and water, and limit the performance of data assimilation systems. To improve accuracy of winter wheat yield estimation, a particle filter algorithm was proposed, which was based on a sequential important sampling procedure of assimilating leaf area index (LAI) and vegetation temperature condition index (VTCI) retrieved from MODIS data into the CERES—Wheat model (Crop environment resource synthesis for wheat) to estimate winter wheat yield from 2008 to 2014 in Guanzhong Plain, Shaanxi, China. In order to determine effects of the assimilated variables on winter wheat yield estimation under different management practices, eight typical rainfed farming sites and four irrigation sites were selected, and the assimilated LAI or VTCI or both of them were used to establish winter wheat yield estimation models. The results showed that the assimilated LAI had good temporal and spatial continuity, and the sharp changing points of seasonal LAI were decreased after applying the particle filter assimilation algorithm. The peak and seasonal trend of the assimilated LAI were basically in agreements with those of the remotely sensed LAI, and the problem of low values of MODIS—LAI was solved to a certain degree after assimilation. The seasonal change of assimilated VTCI was in good agreement with those of both the remotely sensed VTCI and the simulated VTCI, and the assimilated VTCI was a good index for indicating crop water stress of winter wheat. These results suggested that the assimilation of LAI and VTCI might be preferable when the study areas were vulnerable to water stress. At the rainfed farming sites, the determination coefficient of the yield estimation model with assimilated LAI and VTCI was the highest as 0.531 (P<0.001), and the determination coefficients of the yield estimation models with assimilated LAI or VTCI were 0.428 and 0.475, respectively, which were both at the significance level of P<0.001. However, at the irrigation sites the determination coefficient of the yield estimation model with assimilated LAI was the highest as 0.539 (P<0.001), the coefficient of the yield estimation model with assimilated LAI and VTCI was 0.457 (P<0.01), and the coefficient of the yield estimation model with assimilated VTCI was the lowest as 0.243 (P<0.10). In conclusion, the LAI and crop water stress were the important factors that affected winter wheat yield in rainfed farming areas, while the LAI became the important factor in irrigation areas. The study could provide a reference for crop yield estimation by using data assimilation algorithms which combined multi-source remotely sensed variables with crop growth model.