Estimation of distribution algorithms (EDAs) have been developed as a recent kind of evolutionary algorithms during the last fifteen years. Instead of generating individuals through genetic operations (crossover, mutation), they estimate distribution of solutions with higher fitness evaluation: a model of such distribution is constructed and this model is sampled to obtain a new population. In todays EDA, graphical probabilistic and Gaussian models are used most commonly, which are however either computationally infeasible or unrealistic in many real-world problems. Therefore, other kinds of models are appearing. In this paper copulas are used to construct a model of the distribution of feasible solutions. The copula-based EDA (CEDA) is presented with several kinds of copulas, and brief comparison with standard evolutionary algorithms is provided.