The structure of APM is investigated by constraining the loadings of confirmatory factor analysis (CFA) according to item position and item easiness. The constraint of loadings according to an increasing function represents the hypothesis that the item position influences performance in completing APM items. Because of the dependency of item variance on item easiness in binary data this dependency is considered additionally. Several models with one or two latent variables associated with constant and increasing constraints that additionally reflect item easiness were applied to three subsets of APM items. A broad range of item easiness characterized two subsets whereas the range of the remaining subset was rather small. As expected, in the subsets with a broad range the model with two latent variables representing the assumed position effect and dependency did considerably better than the standard CFA model. The superiority of this model suggested that the structure underlying APM is two-dimensional