We report results of the first empirical test, as far as we know, of the assumption of structural invariance of latent constructs from pretest to posttest in cognitive training research on the elderly. In all, 401 participants in the Seattle Longitudinal Study, over 62 years old, received a 5-hr test battery at pre- and posttest that included 16 ability tests, marking the five primary abilities of Spatial Orientation, Inductive Reasoning, Numerical Ability, Verbal Ability, and Perceptual Speed. A total of 229 of our subjects received 5 hr of individual training on either Spatial Orientation or Inductive Reasoning. Restricted factor analysis with the LISREL algorithm tested the hypothesis of measurement equivalence across test occasions, separately for the control subjects and for each of the training groups. When ability-specific cognitive training intervenes, no structural change is observed for abilities not subject to intervention. However, slight shifts occurred in the optimal regression weights for the different markers for the training target abilities.