In this paper we analyze how knowledge spillovers influence the optimal timing of R&D policy. We find that the optimal R&D subsidy profile may be falling, increasing or constant. The time profile of the subsidy rates is crucially dependent on the elasticity of scale in R&D production, i.e. the spillover parameter plus labor elasticity, and the sign of the optimal growth rate of the knowledge stock. Constant subsidy rates are only optimal in knife edge cases. The problems are solved using both an analytical framework and numerical simulations.