New Public Management set off a new wave of performance management efforts in government. Recent performance literature has documented the shortcomings of performance management and provided recommendations for improvement. This emerging-issues article revisits the literature, making a distinction between recommendations for better implementation of what are seen as essentially good systems (single-loop learning) and recommendations that target the performance management systems themselves (double-loop learning). It is argued that in complex settings, performance management may benefit from other ways of doing performance management and needs to be more agile, more decentralized, and more political.