The space of ordered fuzzy numbers (OFN) forms a normed space on which defuzzification functionals can be defined. They play the main role when dealing with fuzzy controllers and fuzzy inference systems. An approximation formula for a general nonlinear functional is given. If a training set is given which describes an action of the functional on OFN then a dedicated evolutionary algorithm can be presented to determine its form. Genotypes composed of chromosomes are proposed together with the fitness function and genetic operators. Some numerical experiments are also performed in the case when ordered fuzzy numbers are given in terms of step functions. For the comparison an approximation procedure with the use of artificial neural networks is also implemented.