Background. Tools chosen to measure poststroke upper-extremity rehabilitation outcomes must match contemporary theoretical expectations of motor deficit and recovery because an assessmentg's theoretical underpinning forms the conceptual basis for interpreting its score. Objective. The purpose of this study was to investigate the theoretical framework of the Wolf Motor Function Test (WMFT) by (1) determining whether all items measured a single underlying trait and (2) examining the congruency between the hypothesized and the empirically determined item difficulty orders. Methods. Confirmatory factor analysis (CFA) and Rasch analysis were applied to existing WMFT Functional Ability Rating Scale data from 189 participants in the EXCITE (Extremity Constraint-Induced Therapy Evaluation) trial. Fit of a 1-factor CFA model (all items) was compared with the fit of a 2-factor CFA model (factors defined according to item object-grasp requirements) with fit indices, model comparison test, and interfactor correlations. Results. One item was missing sufficient data and therefore removed from analysis. CFA fit indices and the model-comparison test suggested that both models fit equally well. The 2-factor model yielded a strong interfactor correlation, and 13 of 14 items fit the Rasch model. The Rasch item difficulty order was consistent with the hypothesized item difficulty order. Conclusion. The results suggest that WMFT items measure a single construct. Furthermore, the results depict an item difficulty hierarchy that may advance the theoretical discussion of the person ability versus task difficulty interaction during stroke recovery. © The Author(s) 2010.