This dissertation examines how firms’ selection of technological and R&D opportunities shape the performance of their innovation efforts. Managers select R&D investments in complex and uncertain environments where it is difficult to learn from past decisions. I examine this challenge using empirical and agent-based modelling methods and by focusing on three interrelated aspects: managers’ individual learning processes, the adaptation of mental representations in complex environments, and the role of distributed expertise in group evaluations. In the first chapter, I propose an alternative explanation to how managers learn from experience that does not involve feedback and that is thus applicable to contexts where learning from feedback is difficult. I test this novel learning mechanism, termed ‘representation learning’, by analysing a large proprietary dataset of patent evaluations and termination decisions made by managers at a Fortune 500 firm. The second chapter explores further implications for performance of representation learning by means of an agent-based model of representation and policy search in rugged landscapes. This study examines how different representation search strategies affect decision-makers’ adaptation in complex environments. Finally, the third chapter explores the performance of group evaluation processes when evaluators differ in the depth and breadth of their knowledge of the technologies being evaluated. This research contributes to management literature by shedding light on the cognitive processes underlying learning and decision-making in uncertain and complex environments. These findings also have practical implications for strategy research and practice concerning the management of uncertain R&D and technology investments.