This article describes a content-focused easy-to-use method of analysing the discourse of Alzheimer's patients. It also reports the results of the method's application to the discourse of seven experimental and four control Alzheimer's patients before and after two semesters of participation in different versions of a multi-modality intervention programme. Eight discourse prompts, representing five different discourse types, were used. Rules for demarcating respondents' transcripts into utterances are presented. Three classes of codes - positive, neutral, and negative - are described, with examples given for each code. Discourse-based outcome measures used were ratio of topic comments to total utterances (TC/U), ratio of different nouns to total nouns (DN/TN), and ratio of vague nouns to total nouns (VN/TN). Other outcome measures were information units (IUs) produced on a picture description task, and scores on a mental status test and a standardised language test battery. All participants received twice weekly physical fitness training and, during the second semester, a weekly session of supervised volunteer work. Experimental participants received, in addition, a prescribed set of memory- and language-stimulation exercises during their fitness workout; control participants experienced unstructured conversation during that time. Interventions were administered by students, supplemented by caregivers. Experimentals outperformed controls on the MMSE and the DN/TN ratio. Neither group declined significantly on the ABCD, TC/U ratio, and VN/TN ratio. Both declined by three IUs on the picture description task, but only the control group's decline was significant. Between-group difference was significant only on the DN/TN ratio. Two experimental participants increased and two had the same MMSE score; three declined. All four controls declined. Discourse assessment is an ecologically valid method of monitoring change in Alzheimer's disease.