Increasing human impact on stream ecosystems has resulted in a growing need for tools helping managers to develop conservations strategies, and environmental monitoring is crucial for this development. This paper describes the development of models predicting the presence of fish assemblages in lowland streams using solely cost-effective GIS-derived land use variables. Three hundred thirty-five stream sites were separated into two groups based on size. Within each group, fish abundance data and cluster analysis were used to determine the composition of fish assemblages. The occurrence of assemblages was predicted using a dataset containing land use variables at three spatial scales (50 m riparian corridor, 500 m riparian corridor and the entire catchment) supplemented by a dataset on in-stream variables. The overall classification success varied between 66.1-81.1% and was only marginally better when using in-stream variables than when applying only GIS variables. Also, the prediction power of a model combining GIS and in-stream variables was only slightly better than prediction based solely on GIS variables. The possibility of obtaining precise predictions without using costly in-stream variables offers great potential in the design of monitoring programmes as the distribution of monitoring sites along a gradient in ecological quality can be done at a low cost.