Background <p>Robot-assisted therapy for poststroke rehabilitation of the upper limb is rapidly spreading. The need comes from the reduced number of therapists and the aim of defining a more involving therapy for the patients. Previous studies report the effectiveness of robotic therapy to provide intensive, repetitive and task-specific rehabilitation, as well as the ability to provide different modalities of training. However, it is not always clear how these modes are implemented and how they are defined, since different labels are sometimes used. As a consequence, it is difficult to define a universal protocol to follow. This leads to non-comparable outcomes, making even harder for the therapists to understand which is the most efficient way of administering therapy. The proposed work aims at putting together available information reported in literature, linking two main variables influencing rehabilitation, i.e., poststroke stage and training modality, with the purpose of updating the state of the art, categorising and analysing the modalities involved, extracting the most effective relationships and approaches in terms of results reported in the scientific community.</p> Methods <p>Scopus was chosen as reference database for the systematic review. The studies refer to the last decade, that is from the year of the latest review published in relation to the topic of interest. Studies clearly referencing training modalities and poststroke stages were included.</p> Results <p>The assistive modality is the one that catches more attention in the scientific community, highlighting the tendency to prefer approaches in which the patients are more actively involved. In terms of relevance inferable from clinical scales reported in the included studies, the assistive modality appears to be the most effective in the chronic phase, active-assistive approaches during the subacute one, whereas no significant conclusions can be drawn for the acute stage. For what concerns robotic devices, some considerations can be drawn as well: exoskeletons are applied during the chronic phase predominantly, whereas end-effectors during the subacute one. No significant distinctions are detected in the acute stage. Improvements in ADLs are mostly achieved in experiments involving exoskeletons, but studies show that subjects may also benefit from end-effectors, when applied in earlier recovery stages.</p> Conclusions <p>It is evident that some connections are present between training modalities and recovery stages, influencing the outcomes of experimental trials. Evaluation metrics exploited in tests report enhanced outcomes when the association between training modality and poststroke stage is optimized. Nevertheless, future developments will possibly extend this study to other factors that may have influenced outcomes, such as intensity of the exercises, frequency and duration of the therapy, and impairment severity. Moreover, a deeper analysis, incorporating investigations on daily clinical practice, would help to identify of the most effective approaches.</p>