Stroke can impair cough function. Respiratory muscle training (RMT) was investigated as an intervention for improving cough function in acute stroke; and as a potential strategy for preventing aspiration-related post-stroke pneumonia. Measures of cough function (volitional tests of cough flow and respiratory muscle strength, automated cough frequency measurement) required validation in the acute stroke population. Test-retest reliability was equally high in eleven healthy volunteers and six stroke survivors (ICCs >0.90). Minimal detectable difference was ≈7%. A calibrated pneumotachograph was found most appropriate for cough flow assessments, due to inaccuracy of portable flow meters (Bland-Altman 95% limits of agreement spanning ≈150 L/min). Automated cough frequency measurements (Leicester Cough Monitor) showed high accuracy (ICC >0.99). The effectiveness of RMT was investigated in a single-blind randomised placebo-controlled trial of 82 acute stroke survivors in three parallel groups (inspiratory, expiratory, and sham training). Mean group changes from baseline (SEM), respectively, were: 91 (42), 49 (27) and 84 (34) L/min for peak voluntary cough flow (p=0.46); -4 (28), 17 (19) and 32 (18) L/min for peak reflex cough flow (p=0.41); 20 (4), 12 (3) and 12 (4) cmH2O for maximal expiratory mouth pressure (p=0.35); and 18 (4), 10 (3) and 14 (3) cmH2O for maximal inspiratory mouth pressure (p=0.40). Pneumonia occurred in 13 (16%) participants with no difference between groups (p=0.65). Higher voluntary cough flow at baseline predicted lower pneumonia risk in patients with unsafe swallow (OR 0.73, 95%CI 0.51-0.95, p=0.012), but not in patients with safe swallow. In a sub-group of 21 patients, 24-hour cough frequency was abnormally high at baseline (median (range) 118 (4, 375)) and decreased to 56 (1, 186) at four weeks and 34 (6, 108) at twelve weeks (p=0.0003). RMT did not improve cough flow or respiratory muscle strength beyond natural recovery. Stronger cough was protective from aspiration-related post-stroke pneumonia.