We explore the historical relationship between financial conditions and real economic growth for quarterly U.S. data from 1875 to 2017 with a flexible empirical copula modelling methodology. We compare specifications with both linear and non-linear dependence, and with both Gaussian and non-Gaussian marginal distributions. Our results indicate strong statistical support for models that are both non-Gaussian and nonlinear for our historical data, with considerable heterogeneity across sub-samples. We demonstrate that ignoring the contribution of financial conditions typically understates the conditional downside risks to economic growth in crises. For example, accounting for financial conditions more than doubles the probability of negative growth in the year following the 1929 stock market crash.