The Sieves method, in statistics, consists in extending a model progressively, as new data are made available. Typically, parameters are progressively added in a statistical estimation method while new samples are provided. We propose an adaptation of the Sieves method in optimization. Decision variables are progressively added while new fitness evaluations are received. We experiment the method on a simple set of noisy optimization problems, and then on a fuzzy control problem applied to unit commitment. The obtained algorithm is simple, applicable to various optimization algorithms (not only evolutionary optimization), and seemingly robust.