The evaluation of labor market policies has become an important issue in many European countries. In recent years, a number of them have opened their administrative databases for evaluation studies. The advantages of administrative data are straightforward: they are accurate, contain a large number of observations (in some cases the whole population) and usually cover a long period of time. However, the information contained in administrative data is normally limited to administrative purposes. Therefore, information that might be relevant for economic modeling is often absent. The IZA Evaluation Dataset aims to overcome such limitations for Germany by complementing administrative data from the Federal Employment Agency with innovative survey data. The administrative part of the dataset consists of a large random sample of inflows into unemployment in Germany from 2001 to 2008 and contains around 920,000 individuals. The complementary survey covers a panel of more than 17,000 individuals who entered unemployment between June 2007 and May 2008. They were initially interviewed shortly after becoming unemployed and then again one year later. In addition, a quarter of individuals were interviewed already after six months. The survey data also contain information on search behavior, ethnic and social networks, psychological factors, (non-)cognitive abilities, and attitudes. This paper describes the sampling and contents of the IZA Evaluation Dataset and outlines the future development.