In order to forecast the content of fumonisin in corn using the infrared spectrum analysis technology, and reduce the differences caused by their yield region, the influence of experiment using 4 different origin of domestic corn were investigated. The method of using x－y co-occurrence distance could be divided into calibration sample and validation sets, using the classical and different regions based on the partial least squares, and then the prediction model of fumonisin maize hybrid origin, and USES the validation set samples to validate the prediction precision, respectively. In order to reduce the computational complexity of modeling and forecasting process, experiments using continuous projection algorithm (SPA) and competitive adaptive weighting algorithm (CARS) the characteristics of the infrared spectra of different origin corn wavelength filter, and 22 characteristics were filtered out. Then these 22 wavelengths were input as variables, and this greatly reduced the computational complexity of modeling and forecasting process, as well as improved the prediction accuracy, with the correlation coefficient at 0.954.