This paper examines problems in the interpretation of attitude data which may arise when the dimensionality of the instrument and sources of heterogeneity in the population are ignored. As a specific example, the paper describes the development and analysis of an instrument designed to measure primary school children's liking of science-related activities. Careful analysis of the patterns of response to the questionnaire created some disquiet about the conventional methods used to collapse data from attitude scales. The results demonstrate that conclusions drawn from attitude measurement can be misleading when proper attention is not given to the structure of the instrument nor to the structure of the population from which the measurements are made.