Despite of its high relevance developmental neurotoxicity (DNT) is one of the least studied forms of toxicity. It is also particularly difficult to analyse, because DNT can be induced by numerous toxic pathways which do not necessarily include cell killing. To bridge this gap we have established five human embryonic stem cell (hESC) based test systems that recapitulate in vitro early neuronal development, including neurulation, patterning, neurogenesis and gliogenesis with in vivo like waves of gene expression. Using the DNT positive control compounds valproic acid (VPA) and methyl mercury (MeHg) we exposed the test systems at the EC10 and at lower concentrations to address the following fundamental questions: 1) Does gene array analysis of the test systems allow differentiation between both DNT compounds? If yes, 2) how large is the overlap between the available ESC based test systems and does analysis of enriched transcription factor binding sites (TFBS) offer advantages over analysis of individual probesets? 3) How can batch effects be controlled? 4) How many independent experiments and how many chips are needed? 4) At which concentrations should gene array analyses be performed? We report that VPA and MeHg clearly induce different gene expression patterns. A much higher number of probesets was deregulated after exposure to the EC10 of VPA (12,996 probesets) compared to MeHg (420 probesets) using the FDR adjusted LIMMA F-test. Relatively little overlap of usually less than 20% was observed between the deregulated probesets in different cell systems, even if all were derived from H9 hESC. However, using TFBS enrichment a relatively large ‘common TF response’ with transcription factors involved into the response to VPA and MeHg could be differentiated from ‘compound specific’ enriched TFs. Matched solvent controls must be integrated into each individual experiment to control for batch effects. To further reduce the influence of batch effects the analysis can be reduced to the 500 probesets with highest variability. Permutation analyses of the results of five independent experiments demonstrates that the number of significant probesets increases with the number of experiments, but if the analysis is restricted to probesets with at least 2-fold alterations also four independent analyses seem to be sufficient for identification of the key genes [besser weglassen?]. The EC10 seems to be an adequate concentration for this type of gene array analysis. In conclusion, the novel ESNATS cell systems offer a basis for the establishment of transcriptomic classification algorithms of DNT.