The ArmAssist is a low-cost robotic system for post-stroke upper-limb telerehabilitation based on serious games. The system incorporates a set of games for the assessment of arm function, the ArmAssist Assessment (AAA), which allows a remote monitoring of the progress of the patient and an automatic adaptation of the therapy. In this study, different components of the AAA are compared against three widely-used clinical tests, the Fugl-Meyer Assessment (FMA) [1], the Action Research Arm Test (ARAT) [2] and the Wolf Motor Function Test (WMFT) [3] in order to select the most clinically meaningful ones for the final score provided to patients and therapist, and evaluate their capability to predict or even improve some aspects of these standard scales. All four tests were performed in 38 separate sessions in 19 post-stroke individuals in their sub-acute phase, as part of a broader study. Statistically significant correlation could be shown with the three clinical tests. These preliminary results are promising for the validation of AAA as a fast, automatic and clinically meaningful tool for remote progress assessment and therapy adaptation; however, more data and further analysis is needed to confirm this.