Abstract: The reactor core design involves search and detailed calculation of a large number of schemes. Shortening the search time is beneficial to improve the efficiency of nuclear design. Through learning and pattern recognition of large data, data mining technology can realize rapid prediction of physical parameters for nuclear design schemes, and fast screening of feasible alternative core schemes. Four decision tree algorithms were used in this paper, C4.5, RepTree, Random Forest and Random Tree, respectively. Fuel enrichment, and the number of fuel rods containing burnable poison and content of burnable poison were taken as independent variables in the calculation. The number of core parameters fulfilled (CPF) was a combination of the four parameters, k_eff unevenness coefficient deviation during lifetime (KUCD), radial power non-uniformity coefficient deviation (RPNCD), radial flux non-uniformity coefficient deviation (RFNCD) and core life (CL). Data mining models were constructed through learning the training set consisted by large number of assembly and core schemes whose nuclear design parameters were already known. Using the models, the CPF values for unknown assembly set (the testing set) were quickly predicted. The results show that model learning time of the four data mining algorithms are less than 0.6 s for the training set, the crossvalidation accuracy of each algorithm is above 0.7 and C4.5 algorithm has the highest accuracy for the overall prediction of CPF values. In addition, for CPF value prediction of testing set, time for each scheme is within 0.9 s, while Random Forest algorithm has the highest prediction accuracy for CPF=4.