Population-based metaheuristics, mostly inspired by biological or social phenomena, belong to a widely used class of approaches suitable for solving complex hard optimization problems. Their effectiveness has been confirmed for many real-time instances of different optimization problems. This paper proposes an Agent-Based Cooperative Population Learning Algorithm for the Vehicle Routing Problem with Time Windows, where the search for solutions is divided into stages, and different learning/improvement procedures are used at each stage. These procedures are based on a set of heuristics (represented as software agents) which are run under the cooperation schemma defined separately for each stage. Computational experiment, which has been carried out, confirmed the effectiveness of the proposed approach.