This paper introduces a new design method of combinational circuits based on genetic programming (GP). An embryo circuit, which combines all evolvable functional cells together, is designed. Each evolvable cell is evolved separately which is the core of the embryo circuit and the whole circuit is designed after the cells evolutionary design finishes. GP is the evolutionary algorithms used to generate a satisfying circuit. The individuals of GP are represented by multi sub-tree groups in order to match the circuits' structures. Corresponding genetic operations are established, including improved crossover and mutation operators. In addition, the paper introduces a random individual set in order to improve the quality of a population. Finally the ability of this improved algorithm for finding optimal solution and its convergence speed are improved much. Final evolutionary results are structures of electronic circuits, which are easily understandable. The paper introduces a method of fitness evaluation based on true tables. The evolutionary results are inputted into MaxplusII10.2 to simulate their functions. In the experiment, two 2 bits ALU is designed, their simulation results shows the method is practicable and the designed circuits are independent on priori knowledge and their functions are satisfying. The comparison experiments proved the presented improved GP speeded up the convergence and improved the ability of GP for finding optimal solution.