Ecotoxicological risk assessment must be undertaken before a chemical can be deemed safe for application. The assessment is based on three components: hazard assessment, exposure assessment and risk characterisation. The latter is a combination of the former two. One standard approach is based on the deterministic comparison of exposure concentration estimates to the concentration of the toxicant below which adverse effects are unlikely to occur to the potentially exposed ecological assemblage. This concentration is known as the ‘predicted no effect concentration’ (PNEC). At the level of hazard assessment we are concerned with, there is a requirement that procedures be straightforward and efficient, as well as being transparent. The PNEC is in general currently determined using either a fixed assessment factor applied to a summary statistic of observed laboratory derived toxicity data, or as a percentile of a distribution over the ecological community sensitivity. Often it is the situation that a hazard assessment will be based on substantially small samples of data. In this thesis we evaluate proposals for determining a PNEC according to regulatory guidance and scientific literature. In particular, we explore these methods under the context of alternative probabilistic models. We also focus on the determination of conservative probabilistic estimators, which may be appropriate for this level of risk assessment. Additionally, we also discuss the detection of species non- exchangeability, a concept which is recognised by scientists and risk assessors, yet typically discounted in practice. A proposal on incorporating knowledge of a non-exchangeable species for probabilistic estimators is discussed and evaluated. The final topic of research examines a generalised deterministic estimator proposed in a recent European Food Safety Agency report. In particular, we analyse the robustness and analytical properties of some cases of this estimator which (at least) maintains the expected level of protection currently attributed. Proposals made within this thesis, many of which extend upon what is currently scientifically accepted, satisfy the requirements of being tractably straightforward to apply and are scientifically defensible. This will appeal to end users and increase the chances of gaining regulatory acceptance. All developments are fully illustrated with real-life examples.