A task was developed to differentiate the reasoning of elementary school students in fifth, sixth, and seventh grades. Subjects completed 16 matrices containing two, three, or four dimensions and varying in the embeddedness of the dimensions. They also solved a formal operational task assessing ability to isolate, exclude, and include variables. Participants were told to look at the eight filled cells of each matrix in order to determine the missing ninth picture. Drawings of the figure which subjects thought completed each matrix were scored for the proportion of dimensions differing from those of the correct figure. Error scores revealed that fifth grade students made more errors across all types of matrices. Only seventh grade students made no more errors of four-dimensional figures than they did on figures with three dimensions. Inclusion of interactions on the formal operational task correlated with error scores on two- and three-dimensional matrices, but not with matrices having four dimensions. It is concluded that results confirm the hypothesis that matrix problem solving continues to develop during pre- and early adolescence and is related to the emergence of formal operational thought. (RH)