Cognitive tasks can be analyzed into cognitive components by using a subtask design, with the subtasks representing the components. The multicomponent latent trait model (MLTM; Embretson, 1980, 1984) was used here for three different tasks: providing a synonym for a stimulus word (composite task), and generating and evaluating synonym candidates (subtasks). The MLTM did not fit the synonym task data. Therefore, a heuristic evaluation procedure was developed that tested the important features of the model. This procedure provided explanations for why the MLTM did not fit the data and, more importantly, which other new models would fit better.