Defining reference condition for stream bioassessment has challenged ecologists for over three decades. Criticisms of the commonly used reference condition approach include: (i) equivocal definitions of ‘minimally disturbed’ reference condition and idiosyncratic approaches to site selection among stream biomonitoring practitioners; and (ii) contending with highly modified areas where near-pristine reference condition does not exist, leading to management decisions based on a shifting benchmark. We used generalised dissimilarity modelling (GDM) to classify stream types based on fish assemblage-environment relationships and objectively define candidate reference conditions for bioassessment. Our study focused on coastal river basins of eastern Australia encompassing the Ecosystem Health Monitoring Program (EHMP) assessment area; a highly modified region with a complex biogeographic history. GDM was implemented using fish presence-absence data at 396 sites linked to functionally relevant GIS-based natural and anthropogenic predictor variables. Stream segments were classified into ecoregions using the GDM-transformed natural variables. Function plots of the retained anthropogenic variables were used to inform the criteria for defining reference conditions. Nine predictors representing multiple spatial scales explained 33.1% of the deviance. Key drivers of species turnover were elevation, air temperature, % unconsolidated rock geology in the upstream catchment, and mean annual runoff. Ecoregions present in the assessment area were also represented in adjacent and less anthropogenically influenced basins to the north and south, and thus amenable to spatial extrapolation of future assemblage-level modelling. This talk will conclude by proposing a framework for objectively and remotely selecting taxa-specific reference condition using GDM and a stratified, probabilistic sampling design.