This paper studies failure to R&D cooperation embracing all research ties, those carried out with research organizations (universities, laboratories and research centers) and/or industrial partners, controlling for firms’ characteristics and sample selection thanks to the French Community Innovation Survey. Using a cross-section model on 2375 innovating firms, we address two questions that are still of interest thanks to the scarcity of empirical evidences: the first one is with whom an R&D cooperation failure is more likely to occur. Secondly, is there any learning effect emerging from previous collaboration? Findings show that failures are more likely to occur when partners are customers, public research organizations or competitors. The likelihood to fail is high and similar for R&D partnerships with competitors or PROs. Further investigation suggests that foreign partners are harder to deals with than national ones and those difficulties are centered on R&D cooperation with foreign universities. Regarding learning to cooperate, our results suggest that former R&D links do not lower the probability to fail with a category of partner even if it boosts the likelihood to cooperate. A stronger appropriation regime influences the likelihood to R&D cooperate but also the probability not to fail. We also provide evidences that large firms, especially belonging to groups, and R&D intensive firms are more likely not to encounter a failure in their R&D alliances. Government support is also found negatively correlated with failure likelihood. Firms with activities in high tech industries or with strong appropriation regimes are also found less likely to fail in their R&D partnerships. Finally, we show that cooperation failures cannot be considered independent from the decision to collaborate in R&D.