Differentiation of cognitive abilities at different levels of general ability in same-aged subjects was tested by a latent variable approach in a multiple-groups design. Eighteen-year old conscripts (n = 14,720) representing the whole range of general ability in the male population, except for the mentally retarded, were tested by the Swedish Enlistment Battery. The sample of subjects was divided into 8, 16 and 32 groups according to level of general ability (G), and the amount of variance due to the ability factors Crystallized intelligence (Gc) and General visualization (Gv) was studied over the G levels. Variance contributed by Gc and Gv increased as the level of G increased. This effect was, however, not observed up through the highest levels of G, probably due to deficiencies in the measurement properties of the Gc and Gv tests at the highest G levels. Theoretical and methodological implications of the results are discussed, as are the consequences for the assessment of specific cognitive abilities at different levels of general ability.