This investigation examined the effectiveness of training Japanese and American children with a visual animation/construction or verbal labelling strategy for solving computerized geometric matrices that were correctly or incorrectly completed and varied in number of elements (l to 3) and number of transformations (0 to 2). Subjects were 209 Japanese and American second, fifth, and eighth graders from top academic schools. Pretests and posttests measured spatial reasoning. Japanese and American children achieved comparable accuracy rates on the test of 108 geometric matrices, but Japanese children had faster response times. Training effectiveness was mediated by gender. Boys benefitted from visual training and girls from verbal training (both, specifically, on incorrectly completed matrices). Cultural and schooling issues helped explain additional gender differences in the American sample. Both cultural groups exhibited "self-terminating" strategies on certain matrices and similar developmental trends. (Author)