Multielectrode array (MEA) neurochips were used to record the electrical activity patterns of neuronal networks in vitro. Seven compounds with pharmacologically relevant targets in the brain were tested to examine the suitability of the system for drug screening: fluvoxamine, fluoxetine (both selective serotonin inhibitors, SSRIs), baclofen, SCH50911, phenobarbital, cortisol, and agmatine. Their effects on the complex activity patterns of the neuronal networks were described with a large number of parameters. Effects of the compounds were different, compatible with the fact that they act on different target systems, while the SSRIs were very similar. It was possible to distinguish the seven substances using machine learning. Effective concentration ranges were similar to therapeutic concentrations and literature data. It is concluded that neuronal networks on MEAs can be applied to characterise compounds, to detect similarities of action, and potentially to identify compounds and their mechanisms of action.