Crop regional simulation has emerged as a new scope for crop model application. It has been used in studies of climate change impact assessment, precision farming, food security and agricultural policy assessment etc. Key obstacles keeping the regional simulation from widely application are availability and quality issues of high resolution daily weather data, and, the diverse methods for generating high-resolution daily weather data from coarse weather observations. Various methods exist in interpolating from randomly distributed weather observation sites to high resolution grid weather data, but how the methods affect the crop regional simulation is rarely known. Evaluating the sensitivity and uncertainty of simulation to method of weather interpolation can help us to identify an appropriate interpolation method for the regional simulation. Based on observed daily weather data from 650 weather observation sites distributed across China, we use three types of interpolation approach, geometrical (choose method of Nearest Neighbor, NN), geographical statistic (choose method of Bivariate Interpolation, BI), and regional climate model interpolation (We use PRECIS Baseline run, BS) to generate gridded (50 km×50 km) daily weather data for whole China, and input them into CERES-Maize crop model. Maize yield is simulated from 1961-1990 and its spatial variability is generated with each interpolation method. Differences in results due to various interpolation methods are measured through (1) comparison of simulated yields to census yields, (2) identifying sensitivity of simulated yields to various interpolation approaches. The census yields from 1980 to 1990 are compared to corresponding simulation yields with each daily weather dataset as input. The comparisons demonstrate interpolated daily weather data with different methods are all able to produce reasonable projections in terms of spatial patterns of yield variability. The spatial patterns simulated by inputting the three interpolation methods are roughly identical to census one, indicating the reliability of the interpolation methods for crop simulation use. Simulated yields are correlate to census yield significantly (P<0.05) in all cases, suggesting the feasibility of using interpolated weather to replace observed weather if observations were not available. Difference exits between census yields and simulated yields, the differences due to different selection on interpolated weather data are within 8%, implying the limited impacts caused by different weather interpolation methods. Sensitivity analysis is operated through correlation analysis for any two of the three simulation results, it proves that there are significant correlations between any two of the three simulation results, but statistically speaking, yields/phonologies are different when comparing any two pairs within the three simulation results. These differences are also significant for most of the maize planting regions. This highlights that caution must be taken before choosing interpolation method for regional crop simulation, particularly in the case of forecasting exact local yield. We make recommendations for selection of interpolation method for crop regional simulation. According to their different characteristics of the methods and the observation data availability, geometrical interpolation is a best solution given the availability and accessibility of nicely distributed and large number of observed weather site, geographical statistic interpolation can be used if regional simulation happens in large flat regions, interpolation by regional climate model is an alternative when attentions were put on spatial variability or without observations.