This paper concerns results of a Rasch and a factor analysis of Raven's Coloured Progressive Matrices items for a sample of 166 gifted children. Of 36 items, 33 had adequate fit to a Rasch model but a three-factor solution provided the most interpretable explanation of item intercorrelations.Raven's Coloured Progressive Matrices (6) was developed as part of a series of nonverbal tests for measuring fluid intelligence ("g") or a person's capacity to reason by analogy, form comparisons, and think logically. Raven's Matrices was developed for use with children ages 5 to 12, aged persons, and mentally retarded persons. Extrapolated norms are provided for children younger than five years. (The Standard Progressive Matrices and Advanced Progressive Matrices are other forms of the test used with other populations.)Raven's Matrices has been held to assess a unitary trait, although Raven suggested five principles underlie problem solution on the Standard Progressive Matrices. These principles are pattern completion, completion of analogy, systematic pattern completion, systematic permutation, and systematic resolution of figures into parts. A number of factor analyses have yielded three or four factors rather than a dominant single factor (1, 2, 9, 10). Among these studies are several in which factor analyses on corrected item intercorrelations are reported. This was deemed preferable since the value of the phi coefficient (a correlation between two dichotomous items) is dependent on item difficulty. The correction removes the effects of this dependency. But, the results of these analyses are conflicting. Rost and Gebert (8) found the correction reduced the number of factors to one while Carlson and Jenson (1) found the correction had little effect and still suggested a three-factor solution. All of the samples used in these investigations came from populations of normal or educable mentally retarded children.The reliability and validity of Raven's Matrices have also been investigated. Marx ( 5 ) found the test had unsatisfactory reliability and an inappropriate spread of item difficulties. Other studies have shown test-re-