Independence after stroke depends on the recovery of motor function, but this is difficult to accurately predict for individual patients. We have previously described an algorithm for predicting potential for recovery of upper limb function for individual patients after stroke. The Predict REcovery Potential (PREP) algorithm begins with a bedside assessment of paretic shoulder abduction and finger extension strength (SAFE score out of 10), followed by transcranial magnetic stimulation (TMS), and magnetic resonance imaging (MRI), as required. Patients are predicted to have potential for an Excellent, Good, Limited or Poor recovery of upper limb function within 12 weeks. The algorithm was developed with data from 40 patients with first-ever ischaemic stroke. This study evaluated and refined the algorithm with a larger, more heterogeneous cohort. Inclusion criteria were confirmed stroke (ischaemic or haemorrhagic), new upper limb motor symptoms, and age at least 18 years. Previous stroke, thrombolysis and thrombectomy were allowed. Exclusion criteria were cerebellar stroke, contraindications to TMS and MRI for those patients who required these tests, and reduced capacity for consent. The Action Research Arm Test was used to measure paretic upper limb function 12 weeks post-stroke. A sample of 192 patients was recruited within 3 days of stroke (106 men, mean age 72 y, 100 right hemisphere), and 157 patients completed the 12 week assessment. The algorithm was refined by combining the SAFE score with age (<80, ≥80 years) to more accurately distinguish between patients with an Excellent or Good prognosis; and by revising the MRI asymmetry index threshold from 0.15 to 0.125 to more accurately distinguish between patients with a Limited or Poor prognosis. These revisions improved accuracy from 59% to 75%. With the revised algorithm, the proportion of patients who need TMS is reduced from over half to around one third. The revised algorithm is therefore more accurate and more efficient. Alternative versions of the PREP2 algorithm will also be presented, which can be used when TMS and/or MRI are unavailable. The potential clinical and economic benefits of implementing the PREP2 algorithm will be discussed.