An enduring belief among management scholars and managers is that unleashing the low-level members of an organization to explore extensively will broaden the exploration conducted by the organization as a whole. Using an agent-based simulation model, we show that in multi-level organizations, increased exploration at lower levels can backfire, reducing overall exploration and diminishing organizational performance in environments that require broad search. Tthis result arises when interdependencies cut across the domains of low-level department managers. In the absence of cross-department interdependencies, more extensive exploration at low levels can improve the performance of the firm as a whole. Our findings show that careful attention to information processing in multi-level organizations can shed light on whether, and when, decentralization encourages innovation.