Entropy-Boltzmann samplings for genetic algorithms are proposed. This selection method is based on the entropy and importance sampling methods in Monte Carlo simulation often used in statistical physics. With the selection methods, the algorithm can explore as many configurations as possible while exploiting better configurations, consequently helping to solve complex optimization problems. To test the performance of the selection method, we adopt the NK-model and compare the performance of the proposed selection scheme with that of canonical genetic algorithms. It is found that the proposed selection method helps to escape local optima and yields a better result. The characteristics of this selection method are discussed in terms of the power spectrum and other analysis.