This paper applies the method of mechanism design to find optimal linear pension rules (contribution rate and monthly benefit function) for flexible retirement: First the government announces a rule, making the benefit dependent on employment length. Each individual, having private information on his own expected lifespan and utility function, optimizes his employment length, conditional on that rule. The government chooses the optimal Bayesian linear rule, which maximizes the social welfare (e.g. the aggregate individual maxima) under a social constraint (e.g. the aggregate net lifetime contribution equals zero). Under this rule there is a better compromise between incentives and insurance than under so-called actuarially fair benefits.