Credit scoring has become a widespread tool to assess the creditworthiness of prospective borrowers, and has been found to increase efficiency and welfare in many settings. But this paper identifies a shortcoming in existing credit scoring systems that may lead to a market failure in agricultural lending in developing countries: Farmers' scores - and their access to credit - decline because of exogenous short-term weather shocks that do not reduce their likelihood of future repayment. I use data on the near universe of formal agricultural loans for coffee production in Colombia to show that excess rainfall shocks cause lower concurrent loan repayment, lower credit scores, and more frequent denial of subsequent loan applications. Drawing on the agronomic literature on coffee production and using survey data, I show that productivity, income and repayment behavior recover faster from these shocks than farmers' credit histories. The additional loan denials create costs for both farmers and the lender that could be avoided. The results suggest that incorporating verifiable information on individual level shocks into credit scores would increase the efficiency of credit markets.