In response to increasing community concern with declines in the ability of aquatic ecosystems to deliver critical goods, services and long-term benefits to human society, there is greater recognition of the need for quantitative procedures to assess aquatic ecosystem health and monitor biotic responses to remedial management. This thesis aims to evaluate the potential to incorporate attributes of freshwater fish assemblages into an ecosystem health monitoring program for wadeable rivers and streams in coastal catchments of south-eastern Queensland, Australia. I identify five key requirements of a quantitative and defensible river health assessment program that need to be evaluated before indicators based on fish can be validly applied for river health assessment in the region. The five requirements are: I) quantification of error associated with sampling fish; 2) assessment of natural ranges of spatial and temporal variation in fish assemblage attributes; 3) accurate definition of the reference condition expected for these attributes in the absence of human disturbance; 4) demonstrated relationships of the indicators with disturbance; and 5) evaluation of potentially important confounding environmental and biological factors. These are critical considerations for minimising the frequency of Type I errors (incorrectly c1assifying a site as impaired) and Type IT errors (incorrectly classifying a site as unimpaired) and accurately assessing river health. The ability to develop an efficient data sampling program without sacrificing accuracy and precision, and hence ability to detect changes through space and time, is a critical requirement of every river health assessment program. I demonstrated that accurate and precise reach-scale estimates of fish assemblage attributes, such as species richness, species composition and species relative abundances, could be obtained from multiple-pass electrofishing plus seine netting of three mesohabitat units and concluded that this was generally a more efficient sampling protocol than less intensive sampling over larger spatial scales. I demonstrated that relatively small differences (e.g. less than 20%) in these assemblage attributes could be detected with a high statistical power (1 -f3 to 0.95) and that relatively few stream reaches (e.g. less than 4) needed to be sampled to accurately estimate assemblage attributes within close proximity (e.g. 20% of the half width of the confidence interval) to the true population means. Fish assemblage attributes that respond to anthropogenic disturbances but exhibit low natural temporal variability are potentially the most sensitive yet robust indicators of human impacts for use in bioassessment programs. I showed that some characteristics of fish assemblages (e.g. species richness, composition and to a lesser extent, species relative abundances) were generally highly stable through time, both seasonally and inter-annually. In contrast, fish assemblages occurring in streams with highly variable flow regimes were more variable. This variability appeared due to natural impacts associated with low flow disturbance. However, fish assemblages at these sites appeared resilient to these natural disturbances, provided that flow and habitat conditions resembled the pre-disturbance state. A critical underpinning of bioassessment programs is the ability to accurately define attributes of the assemblage at a location that are expected in the absence of anthropogenic disturbance (i.e. defining the reference condition). I developed and validated a multivariate predictive model to define the reference condition for native fish species composition based on a large set of least-disturbed reference sites. I also compared four separate approaches to defining the reference condition for native species richness, a univariate biotic attribute underpinning many river health assessment programs. I demonstrated in that spatial variation in these attributes of fish assemblages could be predicted using a small number of simple environmental variables, but that the accuracy of predictions varied with the method used. The multivariate model developed for sites sampled on one occasion and in one season only (winter), was able to accurately predict fish assemblage composition at sites sampled during other seasons and years, provided that they were not subject to unusually extreme environmental conditions. I showed that the predictive ability of the 'maximum species richness line' (MSRL), an approach commonly used for defining the reference condition for univariate attributes such as native species richness, is inherently compromised by the scoring procedure used and because it is limited to predicting variation in the dependent variable along a single environmental gradient (e.g. stream size). This resulted in substantially higher prediction error in comparison to three alternative regression methods, which incorporate single or multiple environmental variables as predictors. The MSRL approach led to an increased frequency of Type I errors for least-disturbed reference and validation sites (i.e. incorrectly classing these sites as disturbed) and compromised the ability of the metric to accurately detect a disturbance 'signal' at the set of notionally disturbed test sites. Assessment of the relationships of the fish indicators with human disturbance gradients indicated that streams affected by human activity due to catchment land use and associated local riparian, in-stream habitat and water quality degradation are more likely to be susceptible to invasion by alien fish species and display major differences in native fish assemblage composition and native species richness from that expected by comparison with similar reference areas minimally affected by human disturbances. Provided that the key requirements of river health assessment identified in this thesis are satisfied, I conclude that indicators based on native fish assemblage composition, native species richness and alien fish species are potentially powerftd indicators of human disturbance and can form the basis for a river health monitoring program in southeastern Queensland and other similarly variable environments in Australia and elsewhere.