Arguments for the validity of animal models generally rely on Causal Analogical Models (CAMs), which are underpinned by Causal Analogical Reasoning. Such arguments were pioneered by Claude Bernard, the 19th century physiologist hailed as the father of experimental medicine, and also known for his fervent anti-evolutionist ideas. Causal analogical reasoning depends on a syllogism: 1. Animal A is similar to animal B, with regards to properties 1, 2, and 3 2. Animal A has property 4 3. Therefore, animal B has property 4. The more similar the properties, the stronger we believe our reasoning to be. In the case of animal models of depression, the argument might look like: 1. A rat is similar to a human, in that under certain conditions, both will show: (1) reduced appetite, (2) decreased activity, and (3) impaired learning/cognition 2. In the rat, these effects can be caused by stressing the organism (4) 3. Therefore, in the human, these effects (symptoms) are caused by stress. In order for such an argument to work, the following characteristics must be true (LaFollette and Shanks, 1996a): 1. The properties (1)–(3) which are common to both animals must be causal properties 2. Properties (1)–(3) must be causally connected with property (4) 3. There must be no causally relevant discrepancy between animals A and B. There is no way to know in advance if characteristic 3 is true, due to the evolutionary differences between species such as rats, mice and humans. It is only true a priori if animals A and B are identical. In an attempt to evade such difficulties, biomedical research tends to employ probabilistic causal reasoning— that is, if a certain percentage of laboratory subjects of species A exhibits a specific phenomenon within the laboratory, then it is inferred that a similar proportion of that species will react in a similar way outside the laboratory. Despite its logical weaknesses, probabilistic reasoning is widespread in medical research and dominates the animal models literature. Most animal models attempt to make predictions in species B (outside of the laboratory) from observations in species A (within the laboratory). As one enters the ‘real world’, it becomes apparent that it is almost impossible to control not only the variables manipulated in the lab, but also the variables that cannot be modelled in animals—cognition, emotion, social behaviour, relationships, etc. Such reasoning is therefore analogical rather than causal (LaFollette and Shanks, 1996b). Basic assumptions about the validity of animal models pervade neuropsychopharmacology, and there are no methodological features which generalise across all experimental techniques (Wright, 2002) The only way to establish whether condition 3 is true (i.e. that there are no causally relevant discrepancies between species or conditions A and B), is to test it empirically. This would require extensive and invasive testing with humans. Unfortunately, the foremost reason for developing animal models is that they negate the need for such testing on humans! It is, therefore, an article of faith that the neural substrates underpinning certain behavioural and cognitive phenomena in experimental animals are the same as in humans. The likelihood of this is arguably greatest for the more ‘primitive’ functions, such as responding for appetitive or aversive stimuli, and least for ‘higher’ aspects of cognition. Brain imaging consistently draws our attention to subregions of the prefrontal cortex as key structures whose altered function is associated with depression (Drevets et al., 1997). How do we model such function in the rat? Although the rat may have a prefrontal cortex, the homology is highly controversial (Preuss, 1995; Brown and Bowman, 2002). Although altered responding for appetitive stimuli in experimental animals has assumed considerable prominence as a potential homologue of human anhedonia, as we have already highlighted above, anhedonia is neither specific to depression, nor invariably present. There seems little prospect of using animals to model those features, which rely on a verbal description of a subjective experience (for example, suicidal ideation, guilt, lack of self-confidence, and low self-esteem). Of course, there are many aspects of cognitive functioning that can be reliably assessed and that appear to be represented in broadly similar ways cross species. Therefore, are there potentially relevant cognitive changes that can be modelled? Regrettably, as above, despite widespread acceptance of their presence and clinical importance, there has been little consistency or clarity in the literature concerning the cognitive effects of depression. In a recent well-designed study of the neuropsychological performance of a group of 44 drug-free patients meeting criteria for a diagnosis of DSM-IV depressive episode, Porter et al. (2003) described widespread impairment across a range of neurocognitive domains. Interestingly, in contrast to previous reports, there was evidence of substantial recognition memory deficits despite intact declarative verbal learning and memory, perhaps pointing towards different questions to be asked of existing animal models?