Latent trait models are presented that can be used for test design in the context of a theory about the variables that underlie task performance. Examples of methods for decomposing and testing hypotheses about the theoretical variables in task performance are given. The methods can be used to determine the processing components that are involved in item performance. Three component latent trait models for underlying theoretical variables are described along with their maximum likelihood estimators. The item parameters can be used for item banking according to the influence of the underlying processing variables on item difficulty. Such estimators permit the test developer to choose items that represent specified information processing demands for the examinee. In this manner, what is measured by an aptitude test can be explicitly designed by specifying difficulty levels in the underlying processing components. The need for meta component latent trait models was also considered. It was shown that both items and persons vary on metacomponent parameters and that these parameters are important for the predictive validity of an aptitude test. The main conclusion to be drawn from these studies is that metacomponent latent trait models are needed to estimate more fully the processing abilities that underlie aptitude. (PN)