Understanding the susceptibility of financial systems to systemic risk, and in particular the contribution of inter-bank credit relationships, has been a key issue post-crisis. With a few notable exceptions, however, it is difficult to identify these relationships. In this paper we analyse systemic risk in the US banking system before, during and after the Panic of 1873. Banks at this time were simpler in their investment activities, making balance sheets more informative, and a combination of regulations and pre-internet technology limited the degree of bank interactions. We use a unique numerical approach to simultaneously estimate the interbank network and the utility functions of banks with pre-crisis data. Using this network we show that direct counterparty risk in this period is small – in agreement with historical observation. Losses from failures are secondary in significance to liquidity shortages caused by panicked deposit withdrawals. Comparison of our network and post-crisis data allows us to explain changes in net balance sheet quantities and to correctly predict which banks panic in response to the crisis. The results increase our understanding of the historical crisis whilst simultaneously suggesting insights into the study of systemic risk in modern financial systems.