Interactive software in general, and virtual environments in particular, have potential as an aid to learning for both children and adults with learning disabilities. More needs to be known, however, about the best way to exploit it. The study was designed to identify what strategies human tutors use when working with adults who were learning to use virtual environments and to investigate their effectiveness by examining changes over time in both tutor and learner. The intention of the study was to produce a repertoire of effective tutor strategies that could be evaluated at a later stage in a systematic intervention study. Three steps were taken to achieve this: reliable categories of tutor and learner behaviour were developed; these categories were examined over time to see whether they differed in frequency of occurrence, in order to validate their differentiation; an attempt over time was made to determine effectiveness by relating occurrence of different tutor behaviours to learners' goal achievement. The dataset consists of five data files: mw2.por contains rates per second of eleven tutor behaviours and positive or negative goals achieved by learners for all sessions, for all learners; crspss.por contains data from the road crossing environment (557 cases); caspss.por contains the data from the virtual cafe (1025 cases); smspss.por contains data from the virtual supermarket (1307 cases); faspss.por contains data from the virtual factory (1379 cases).