The natural defensive behaviors of laboratory mice have been evaluated in both seminatural and highly structured situations; and characterized in terms of eliciting stimuli, response to pharmacological agents, behavior patterns, and outcome or effect on the social and physical environment. The defense patterns of laboratory mice and rats are generally similar, but mice show risk assessment on initial exposure to highly threatening stimuli while rats do not, while rats display alarm vocalizations, missing in mice. Quantitative differences in freezing and¯ight for laboratory mice and rats appear to largely re¯ect domestication effects, with wild mice and rats more similar to each other. This nexus of detailed within-species and comparative data on defense patterns makes it possible to reliably elicit speci®c defenses in mice or rats in an experimental context, providing well-validated assays of the natural defensive behaviors themselves, as opposed tò models' of defense.The mouse±rat comparisons indicate considerable cross-species generality for these defense patterns, as does a scattered but considerable literature on other mammalian species, generally involving ®eld studies and typically focusing on those aspects of defensive behavior that are visible at a distance, such as vigilance, or¯ight. Although potential homologies between normal mouse and human defense systems should ideally involve all four pattern components (stimulus, organismic factors, response characteristics, outcome), predictive validity in terms of response to drugs active against speci®c defensive psychopathology is the most extensively investigated of these. Flight, as measured in the Mouse Defense Test Battery shows a consistently appropriate response to panicolytic, panicogenic, and panic-neutral drugs, while some other predictive`panic models' (dPAG-stimulation; DMH-inhibition; possibly conditioned suppression of drinking paradigms) also elicit and (indirectly) measure behaviors potentially related to¯ight. Models unrelated to¯ight (e.g. ultrasonic vocalization to conditioned stimuli); or for which¯ight elements may a relatively minor contributor to the behavior measured (Elevated T-maze) are less predictive of panicolytic or panicogenic action. These ®ndings indicate that natural defensive behaviors provide a well-characterized pattern for analysis of effects of genetic or other physiological manipulations in the mouse, and may also serve as a model for analysis of defenserelated human psychopathology. q 2001 Elsevier Science Ltd. All rights reserved. The recent explosion of genetic techniques and genetically modi®ed animals has greatly exacerbated the need for comprehensive, natural, models or assays of behavior that are relevant to mice, the mammalian species of choice for genetic research [25]; and has additionally focused attention on issues of construction and validation of these tests.Minimally, a behavioral model/assay should enable the reliable elicitation and measurement of a qualitatively consistent behavior pattern for ...