Abstract: One or more algorithms are used to, in respect of each of a plurality of EEG snippets from a subject, allocate to each snippet a class selected from one or more classes, each class being associated with drug impairment level. The one or more algorithms are each defined by a model and a set of parameters and have been previously trained with a plurality of labelled EEG snippets, the labelled snippets including, for each of the classes, at least one snippet associated with a test subject experiencing the level of drug impairment associated with said each class. A statistical model is created which provides desired levels of sensitivity and accuracy based upon the proportion of snippets from test subjects having the known level of drug effect that falls within the associated class. Drug effect is defined by applying the statistical model to the classes allocated to the snippets.