Defect-free blades with fine microstructure and excellent high temperature performance are needed by aero-turbine. The curved shape of SX (single grain) blade is usually very complicated, and its section varies sharply at the platform, the tip shroud, etc. Thus, stray grains are difficult to avoid during directional solidification. In order to optimize the process parameter, the physical model and the mathematical model of directional solidification of SX blade were established based on Panda thermodynamic database and finite element. The temperature field, the evolution of mushy zone and the SDAS (secondary dendrite arm spacing) during solidification of SX sample by different process were simulated, and the discipline and mechanism of defect formation were investigated. The numerical results agree well with the experimental. The calculated results indicate that with the higher withdrawal rate, the SDAS gets smaller but stray grains tend to occur; with the lower withdrawal rate, the possibility of stray grains gets lower but the SDAS gets larger. The directional solidification of complicated thin-walled hollow real SX blade was simulated as well. The results show that the SDAS is nearly uniform in the blade, stray grains may appear at the platform when the withdrawal rate is up to 3.5 mm/min. By the process with varying withdrawal rate, the stray grain can be avoided while most of secondary dendrite can be refined, thus improving the productivity and percent of pass