Development of in vitro systems, such as those based on embryonic stem cell differentiation, depends on the selection of adequate test and training compounds. We recommend the use of two classes of positive controls, the "gold standard compounds" for which developmental neurotoxicity (DNT) has been proven in man, and the "pathway compounds" that are known to disrupt signalling pathways and key processes relevant for neuronal differentiation. We introduce the concept of toxicity endophenotypes (TEP) as changes in neuronal connectivity resulting from exposure to developmental toxicants. Thus, TEPs provide the scientific rationale for modeling DNT with simple in vitro models of key neurodevelopmental events. In this context, we discuss scientific and technical aspects of the test compound selection process. We suggest to include compounds with unspecific toxicity, besides negative control compounds, and we recommend tandem approaches to determine relative toxicities instead of absolute measures. Finally, we discuss how to avoid pitfalls by distinguishing between unspecific forms of cytotoxicity and specific developmental neurotoxicity. A compilation of compound lists corresponding to the above-discussed principles supplement this review.