Higher-tier tests for the assessment of early treatment effects should be aimed at providing specific information on the behavior processes affected, rather than simply at extending the descriptive data base. The contrast between positive and negative results can be useful to point out possible mechanisms of action. For example, late prenatal oxazepam exposure of mice produced a reduction of the amphetamine hyperactivity at the end of the second postnatal week, but did not significantly affect the response to scopolamine at the end of the third week. An impairment of active locomotor avoidance was observed at the young adult stage, which contrasted with the absence or scarcity of changes in passive avoidance and extinction responding in the same go-no go tests. These changes in response-activating mechanisms appear to be in agreement with the medium- and long-term effects on CNS monoamine metabolism described in the literature. As concerns statistical analysis, dichotomous or polytomous data obtained, e.g., by the Fox battery are not yet amenable to an adequate processing, due to the shortcomings of the available nonparametric tests. By contrast, mixed-model ANOVAs can cope with complex data obtained, e.g., in activity and learning tests. However, the available checks on various assumptions (normality, homogeneity of variance, sphericity) are not valid when nested factors, block factors and repeated measures coexist. Finally, the more usual cross-fostering procedures provide adequate information on some aspects (e.g., separation of main effects of prenatal treatments from postnatal maternal effects) but not on others.(ABSTRACT TRUNCATED AT 250 WORDS)