Evaluation of training effectiveness is a long-standing problem of cognitive intervention research. The interpretation of transfer effects needs to meet two criteria, generality and specificity. We introduce each of the two, and suggest ways of implementing them. First, the scope of the construct of interest (e.g., working memory) defines the expected generality of transfer effects. Given that the constructs of interest are typically defined at the latent level, data analysis should also be conducted at the latent level. Second, transfer should be restricted to measures that are theoretically related to the trained construct. Hence, the construct of interest also determines the specificity of expected training effects; to test for specificity, study designs should aim at convergent and discriminant validity. We evaluate the recent cognitive training literature in relation to both criteria. We conclude that most studies do not use latent factors for transfer assessment, and do not test for convergent and discriminant validity.