Recognising a need for code and data validation in the area of accelerator-driven systems (ADS), in 1999 the OECD/NEA Nuclear Science Committee organised a benchmark on a lead-bismuth cooled subcritical system driven by a beam of 1 GeV protons. Seven organisations (ANL, CIEMAT, KAERI, JAERI, PSIICEA, RIT and SCK-CEN) contributed to this benchmark exercise using different nuclear data libraries based mainly on ENDFIB-VI, JEF-2.2 and JENDL-3.2. For the reactor analysis, both deterministic and Monte Carlo methods were applied. The analysis of the results shows significant discrepancies in important neutronic parameters, such as one-group microscopic crosssections and kerr. arising from differences in the basic nuclear data, the data processing methods and the neutron transport approximations. An experiment-based benchmark exercise may help to resolve the discrepancies observed for the fresh cores.