Companies innovating in dynamic environments face the combined challenge of unforeseeable uncertainty (the inability to recognize the relevant influence variables and their functional relationships; thus, events and actions cannot be planned ahead of time) and high complexity (large number of variables and interactions; this leads to difficulty in assessing optimal actions beforehand). There are two fundamental strategies to manage innovation with unforeseeable uncertainty and complexity: trial and error learning and selectionism. Trial and error learning involves a flexible (unplanned) adjustment of the considered actions and targets to new information about the relevant environment as it emerges. Selectionism involves pursuing several approaches independently of one another and picking the best one ex post. Neither strategy nor project management literatures have compared the relative advantages of the two approaches in the presence of unforeseeable uncertainty and complexity. We build a model of a complex project with unforeseeable uncertainty, simulating problem solving as a local search on a rugged landscape. We compare the project payoff performance under trial and error learning and selectionism, based on a priori identifiable project characteristics: whether unforeseeable uncertainty is present, how high the complexity is, and how much trial and error learning and parallel trials cost. We find that if unforeseeable uncertainty is present and the team cannot run trials in a realistic user environment (indicating the project's true market performance), trial and error learning is preferred over selectionism. Moreover, the presence of unforeseeable uncertainty can reverse an established result from computational optimization: Without unforeseeable uncertainty, the optimal number of parallel trials increases in complexity. But with unforeseeable uncertainty, the optimal number of trials might decrease because the unforeseeable factors make the trials less and less informative as complexity grows.