Sources of item bias located in characteristics of the test item were studied in a reasoning test developed in South Africa. Subjects were 1,056 White, 1,063 Indian, and 1,093 Black students from standard 7 in Afrikaans and English schools. Format and content of the 85-item Reasoning Test were manipulated to obtain information about bias or differential group achievement. Bias indices were calculated by the Scheuneman technique and the three-parameter logistical model. The Junior Aptitude Tests, the Raven Standard Progressive Matrices, and a biographical questionnaire were used, along with the Reasoning Test, in this investigation. Results suggest a distinction between true and apparent item bias. True item bias included aspects such as language factors, overly attractive distractors, and a lack of the knowledge and concepts that are usually acquired informally. Apparent bias--a feature of the examinee's test behavior rather than of the item--was indicated as a tendency to use associative, rather than conceptual similarity in classification, selective attention, and an insufficiently logical approach. Findings also suggest that: (1) items in which a statement was made functioned less well than did other types; (2) story-type items led to poor performance; (3) sayings were not suitable open item types; (4) figures were superior to language in terms of content; (5) figural analogies were superior to verbal analogies; (6) series were better than analogies in terms of format; and (7) syllogisms were not suitable item types. Twenty-seven tables and one figure contain study data. (SLD)