Most existing evolutionary approaches to multiobjective optimization aim at finding an appropriate set of compromise solutions, ideally a subset of the Pareto-optimal set. That means they are solving a set problem where the search space consists of all possible solution sets. Taking this perspective, multiobjective evolutionary algorithms can be regarded as hill-climbers on solution sets: the population is one element of the set search space and selection as well as variation implement a specific type of set mutation operator. Therefore, one may ask whether a 'real' evolutionary algorithm on solution sets can have advantages over the classical single-population approach. This paper investigates this issue; it presents a multi-population multiobjective optimization framework and demonstrates its usefulness on several test problems and a sensor network application.