Rivers and streams are among the most imperilled ecosystems on earth owing to overexploitation; water quality impacts, altered flow regimes, habitat destruction, proliferation of alien species and climate change. There is a pressing need to address these threats through stream bioassessment, stream rehabilitation and species conservation actions. Species distribution models (SDMs) can offer a practical, spatially explicit means to assess the impact of these threats, prioritise stream rehabilitation and direct conservation decisions. However, applications of SDMs for stream bioassessment and real-world conservation outcomes in freshwater ecosystems is still in its infancy. This thesis set out to link conceptual advances in fish ecology with emerging statistical methods applied to stream bioassessment and species conservation issues facing eastern Australian freshwater fish species. One of the primary uses of SDMs in freshwater environments is bioassessment, or assessment of “river health”. A network of reference sites underpins most stream bioassessment programs, however, there is an ongoing challenge of objectively selecting high quality reference sites, particularly in highly modified assessment regions. To address subjectivity associated with ‘best professional judgement’ and similar methods, I developed a novel, data-driven approach using species turnover modelling (generalised dissimilarity modelling) to increase objectivity and transparency in reference site selection. I also tested whether biogeographic legacies of fish assemblages among discrete coastal catchments limited the use of reference sites in southeast Queensland and northeast New South Wales. The data-driven approach was then used to select reference sites and sample fish assemblages to develop freshwater fish SDMs for subsequent data chapters. Another factor potentially limiting the accuracy of SDMs for bioassessment and conservation is the modelling strategy employed. In particular, site-specific models for stream bioassessment usually still use ‘shortcut’ methods such as community classification and discriminant function analysis, despite growing evidence that machine learning algorithms provide greater predictive performance. I tested how reference coastal fish assemblages are structured in relation different species assembly theories (e.g., species arrangement in discrete communities, species sorting independently across environmental gradients, or elements of both) by comparing different modelling approaches reflective of these processes (community level modelling, stacked ‘single species’ models and multi-species response models). Evaluation of the modelling was used to determine which of these modelling paradigms best suit stream bioassessment and other conservation applications such as survey gap analysis, estimating range changes owing to climate or land use change and estimating biodiversity. The taxonomic completeness index is the most commonly used site-specific index for stream bioassessment programs, despite several recognised limitations of this index, including use of an arbitrary threshold; omission of rare taxa that may be responsive to subtle levels of disturbance; and omission of potentially useful information on taxa gained at disturbed sites. I developed and tested an index that incorporated both native species losses, and gains of tolerant and alien species into a unified index of assemblage change for stream bioassessment. This study used a single species ensemble modelling approach to predict species occurrence and combined predictions into an index akin to Bray-Curtis dissimilarity. The resultant index, ‘BCA’, markedly outperformed the widely used taxonomic completeness index derived from community classification (discriminant function analysis) models and has considerable potential for improving stream bioassessment index sensitivities for a range of freshwater indicators (e.g. diatoms, macroinvertebrates, macrophytes). It is recognised that there are very few peer-reviewed SDM studies that have ‘real world’ conservation applications; most are instead academic exercises concerned with addressing methodological challenges, or hypothetical examples of how one might apply a SDM for a conservation problem. To address this gap between modelling and management, I used SDMs to inform a conservation plan for declining southern pygmy perch (Murray-Darling Basin lineage) (Nannoperca australis) in northern Victoria. This study incorporated alien species abundance models as predictors into an ensemble SDM to identify remnant habitats of this declining species. The models indicated that ~ 70% of N. australis habitat has become unsuitable since European settlement owing to anthropogenic pressures and interaction with alien fish species, particularly brown trout (Salmo trutta). Model outputs were used for survey gap analysis and to identify stream segments suitable for targeted management and reintroduction of the species. This study formed the basis for a captive breeding and translocation plan for southern pygmy perch in northern Victoria. The thesis concludes with practical learnings from these modelling studies for freshwater bioassessment and conservation practitioners; namely: (1) that machine learning multispecies response and ensemble models offer improved predictive performance compared with traditional approaches and that model choice depends on the intended use of the model; (2) that a newly developed index, “BCA”, provides a more conceptually sound and sensitive index than the traditionally used taxonomic completeness index for stream bioassessment; and, (3) that SDMs developed using readily available and high quality stream bioassessment datasets provide an excellent foundation for applied freshwater fish species conservation and management. The thesis concludes with future challenges and directions for freshwater fish SDM research.