A default of a bank has cascade-effects in a financial network in which entities are tightly intertwined. The cascade may propagate sequentially with additional defaults, from close neighbors to distant banks. Many contributions show that banking systems seem to be fairly stable to contagion via credit risk, as very large shocks are needed to simulate cascades of a meaningful size. We use a novel method - DebtRank – from previous contributions of one of the authors, to assess the centrality of a bank in a network, accounting for the propagation of distress even in the absence of defaults in the cascade. Indeed, an event that weakens the balance-sheet of a bank j, has a negative spillover on the balance sheet of claim-holders of j (contagion through distress). In this respect, the centrality of a bank is (i) proportional to its relative exposure toward the source of distress and (ii) depends on its financial soundness. DebtRank solves the infinite reverberation problem typical of contagion in networks with loops. We estimate the total potential loss to the financial system caused either by an initial default of a single institution or by a common shock to several institutions. The method also allows to find candidate subsets of institutions that, together, may constitute systemically important groups. We use a unique dataset of supervisory reports to the Bank of Italy that includes (i) bilateral exposures (secured and unsecured, short and long term) between all Italian banks, (ii) the links with major foreign financial institutions and (iii) balance sheet data (capital, total and encumbered assets,…).