The constraints of genetic spectrum assignment model without co-channel interference in cognitive radio networks are analyzed in this paper. An improved genetic spectrum assignment model is proposed, the population of genetic algorithm is composed of sets of feasible assignment and infeasible assignment. The penalty function is added to the fitness function, and the fitness of chromosomes that dissatisfied with the constraints are reduced, thus the feasible assignment is achieved with the objective of maximizing the system utility. The simulation results show that the proposed algorithm can achieve higher system utility with relatively more cognitive users under the condition of same population size and other related parameter.