We investigated mental rotation in children by systematically varying the adult cube aggregate's set size, rotation angle, and picture/depth plane rotations in a new test. Eighty 4- to 11-year-old mainly middle-class children (British Indian and British African majority and white minority; 40 girls and 40 boys) were assessed using the new matching-to-sample Colored Mental Rotation Test (CMRT) and, for comparison, the Raven Colored Progressive Matrices Test (RCPM). A high Cronbach's alpha of 0.94 and the Rasch model demonstrated item homogeneity of the CMRT. As expected, there were main effects of age showing increases in accuracy and of sex as boys outperformed girls. A main effect of set size showed that from age four until age 10 as four-cube aggregates proved to be the most economical three-dimensional (3D) object for mental rotation. Several higher-order interactions all involved four-cube aggregates, for example, 3D cube element protrusions had the largest effect in the four-cube-aggregate. We thus suggest that the magical number four (Cowan, 2001) as an attentional limit may also be valid in mental rotation and linked to the 'Good Gestalt' design of the four-cube aggregate. The cross-validation of CMRT with the RCPM showed high correlations increasing from 0.69 in 4- to 5-year-olds to 0.77 in 10- to 11-year-olds. Interestingly, 4- to 5-year-olds girls scored higher in the Raven test of nonverbal reasoning than in the CRMT scores with 3D cube aggregates demonstrating the particular complexity of 3D pictorial space.