Based on meteorological data of 80 representative meteorological stations, maize yield of 44 counties and maize growth period of 31 agrometeorological stations in Heilongjiang Province from 1981 to 2014. Methods for dynamic prediction of maize in the period of ten days of March to July were established according to principles of integral regression through the expansion of meteorological data. The province was divided into 4 regions as needed.The results showed that: the average accuracy of the back tests was 87.4%. The last in III region was 82.5%. The model average prediction accuracy rates were also 87.4%. The accuracy in I region and II region were 91.5% and 92.0% separately, moreover they were performing better in the two regions than in III region and IV region.