While prior studies have investigated the effect of collaborative R&D with different partner types (suppliers, customers, competitors and research institutions and universities) on firms' innovative performance, the implications of dynamic patterns in these collaborations have not received attention. In a large panel of Spanish innovating firms operating in a broad range of industries during the period 2004-2011, we examine the differential effects of recently formed, persistent, and recently discontinued collaboration on innovative performance. Persistence is the most common pattern of collaboration, while discontinuities are most often observed for competitor collaboration. We find that it is persistent collaboration that has a systematically positive effect on innovativeness. With the exception of recently formed collaboration with universities and research institutes, other temporal patterns of collaboration do not significantly improve innovation performance. Implications of these findings are discussed.