A method to predict grain structure defects in directional solidification precise casting has been developed based on a solidification process simulation with a three-dimensional CAFE (cellular automaton-finite element) model. In our paper we studied which parameters of the dendrite growth model in the CAFE model are important by sensitive analyses. Then, the dendrite growth model was advanced by evaluating quantitatively liquid diffusion coefficients dependent on temperature and liquidus slope dependent on liquid concentrations which have much effect on the growth model using molecular dynamics simulations and Thermo-Calc calculations. The method to forecast grain structure defects has been applied to directional solidification casting of a simple shape and a gas turbine blade, and has been verified with comparison between the simulated results and experimental results on locations or grain directions of defects, solidification microstructure. It was found that our method could predict a stray grain defect in directional solidification precise casting.