A wide range of approaches has been used to study anxiety in mice. All presuppose that aversive stimuli, such as foot shock or novelty, induce a central state of fear, which can be quantified through specific behavioural and physiological measures. This review discusses the validity of the various approaches in terms of their similarity to different human anxiety disorders, their ability to detect compounds which modulate human anxiety, and their relevance to animal defensive processes. The most commonly used models of anxiety suitable for screening transgenic and knockout mice are discussed, with an emphasis placed on controlling for factors which could confound results. As all models used to date have limitations and no single paradigm adequately models all aspects of anxiety, this review recommends the use of a broad range of anxiety models in order to provide a comprehensive characterisation of the behavioural phenotype of transgenic mice.