The Netherlands Environmental Assessment Agency (MNP) attempts to produce policy assessments and evaluations that are both useful for policy and politics and scientifically valid. A 'boundary work' perspective is used to analyse how the interface between science and policy is produced and reproduced discursively. Analysis of the general discourse about the position of the MNP and the relation between science and policy is followed by a case study on the environmental effect assessment of political party election manifestos. Pielke's classification of scientists' roles in policy is used to typify the discourses identified. On a general level, MNP employed 'Science Arbiter' discourses, steering clear of political considerations but trying to meet policymakers' demands for assessment and information. However, in the case study, discourse moved in the direction of the 'issue advocate', with science used to steer policy in a certain direction. The dynamics behind these processes are analysed and their implications discussed.