Decentralised experimentation and mutual learning of public policies is seen as one of the important advantages of federal systems (Oates: laboratory federalism). Based upon Hayekian ideas of the advantages of decentralised experimentation (as a discovery procedure), we analyse the long-term benefits of parallel experimentation in a federal system from an evolutionary economics perspective. We present a simulation model in which the lowerlevel jurisdictions in a federal system experiment with randomly chosen policy innovations and can imitate the relatively best solutions. The simulations confirm our hypotheses that a higher degree of decentralisation has positive effects on the long-term accumulation of knowledge of suitable policy solutions and also limits risks through better protection against erroneous policies. Both problems of policy learning and trade offs with (static and dynamic) advantages of centralisation are taken into account.