Stroke is a heterogeneous disease with risk factors and treatment that depend on the specific etiopathogenetic mechanisms. Stroke is the leading cause of short-term functional impairment in adults, which can lead to significant long-term disability in the absence of adequate treatment and rehabilitation. Recovery from stroke is a complex process with patterns that differ among patient groups. A key challenge in post-stroke rehabilitation and research of new interventions is precisely the variability of the degree of recovery among clinically similar patients, which makes it difficult to assess the effect of a particular rehabilitation method on recovery. Underlying the recovery process is neuroplasticity which forms the basis of spontaneous biological recovery and learning-dependent recovery mechanisms. A better understanding of these mechanisms could help predict recovery potential and plan personalized rehabilitation. To predict stroke recovery as accurately as possible, at the individual level, prediction models across different functional domains are being developed. The models include clinical measures and biomarkers of damage or recovery potential of the nervous system. Biomarkers, which include biological, physiological, and radiological measures, should improve the accuracy of recovery predictions based on clinical measures. Prediction tools should combine a small amount of information in a systematic and easy-to-remember way for easier implementation in clinical practice. Informing patients and their families about the expected recovery potential, as well as possible poor health outcomes, enables better adaptation to a new life situation and future planning.