Zebrafish are increasingly being used in behavioral neuroscience, neuropsychopharmacology and neurotoxicology. Recently, behavioral screens used to model anxiety in rodents were adapted to this species, and novel models which tap on zebrafish behavioral ecology have emerged. However, model building is an arduous task in experimental psychopathology, and a continuous effort to assess the validity of these measurements is being chased among some researchers. To consider a model as valid, it must possess face, predictive and/or construct validity. In this article, we first review some notions of validity, arguing that, at its limit, face and predictive validity reduce to construct validity. Then we review some procedures which have been used to study anxiety, fear or related processes in zebrafish, using the validity framework. We conclude that, although the predictive validity of some of these models is increasingly being met, there is still a long way in reaching the desired level of construct validity. The refinement of models is an ongoing activity, and behavioral validation and parametric research ought to advance that objective.