The lack of bank risk data is one of the major challenges in bank risk management. This paper proposes a general framework for bank risk data set construction, which provides an integrated process from data sources to comprehensive risk data sets. Specifically, from papers, surveys, websites and other open data sources, some data or information on bank risk can be obtained. Then, we categorize this data or information into different types and give corresponding approaches to turn them into comprehensive and standard bank risk data sets. To the best of our knowledge, this is the first paper that tries to offer a feasible solution to the problem of data sparseness in bank risk management. By using the proposed framework, three real-world bank risk data sets are constructed, among which the Chinese banking operational risk data set is the most comprehensive in China, with a total of 2132 risk records and fifteen features to describe every risk event. The Chinese listed banks risk data set contains credit, market and operational risk values for all sixteen listed banks from 2007 to 2014. The Austrian banking risk data set consists of credit, market and operational risk distributions of the entire Austrian banking system in September 2002.