In current work, commercial cerium has been zone refined by induction heating in a cold crucible against the pollution from the crucible materials. A numerical model of solute redistribution during zone refining is integrated into the genetic algorithm (GA) to search after the optimum zone length in different zone passes for increasing the redistribution efficiency of solutes, and the effect of variable distribution coefficient on purification efficiency is also considered in model. The chemical analysis results after 15 zone passes have showed that the content of Fe and Ni at center of sample bar are reduced by 50% and 90%, and the major difference of these two solutes with close effective distribution coefficient is their initial content. The main difference of experimental results and calculation results is from the instability of shorter zone during later zone passes. Experimental and calculation results show that GA is powerful tool for looking for optimum of zone lengths in zone passes, but the stability should be maintained in experiment by effective method to promote smoothly mobility of solutes in molten zone; the initial content of solutes should be kept as low as possible for getting better purification effect.