This paper exposes a fundamental tension between the micro-prudential objective of subjecting banks to greater discipline through debt markets and the macro-prudential objective of containing systemic risk. We start with the familiar result that banks are illiquid due to the inability of bankers to credibly pre-commit to asset choices. Bank debt can reduce this illiquidity by disciplining bankers with the threat of premature liquidations. These liquidations are ex post inefficient in the absence of systematic shocks to asset values. To reduce some of this inefficiency, the discipline of equity can be used by incenting the manager through equity-based compensation. However, this reduces the bank’s liquidity, so some debt is socially optimal in bank capital structures. When we introduce systematic shocks to asset values, another aspect of bank leverage emerges─the liquidation of a bank’s assets leads creditors of other banks to update their priors on these common shock affecting asset values. This can give rise to contagion and liquidations throughout the system that can be either efficient or inefficient. Thus, individual-bank liquidity creation induced by the disciplining role of bank debt has the benefit of generating more information about common asset-value shocks, but it also comes at the cost of greater systemic risk. This increase in systemic risk is not fully internalized by banks in choosing privately optimal levels of leverage. Thus, the leverage that represents the regulatory optimum is lower than each bank’s privately optimal leverage. We then consider two different information regimes for lender of last resort (LOLR) interventions. In the first regime, the LOLR has no information about the idiosyncratic conditions of banks or the systematic asset value shock. In this case, the LOLR can diminish the incidence of contagion and thereby reduce systemic risk by agreeing to bail out liquidation-threatened banks, but this carries the misfortune of eliminating creditor incentives to intervene in banks and liquidate efficiently. The analysis points to the desirability of ex-ante capital requirements as an alternative to ex-post LOLR interventions. In the second regime, we permit the LOLR to learn about the systematic shock and show that now contagion can be eliminated without affecting market discipline by adopting a selective intervention policy.