Background: The complexity of the heart failure with preserved ejection fraction (HFPEF) syndrome makes that the best approach towards diagnosis is still not fully established. According to recent studies stress echocardiography may be useful to improve the diagnosis, but its exact place in clinical practice is still unproven. Methods: Tissue Doppler velocity traces were acquired from 55 subjects at rest and during submaximal exercise (semi- supine bicycle) at the basal septum and lateral wall, and temporally aligned to a common reference for comparison. This population consisted of 19 HFPEF patients (diagnosed by the HFA of ESC criteria), 14 controls with unexplained breathlessness (EF>50%, did not fulfill HFA criteria) and 22 healthy controls. Each stage of the stress protocol was independently analyzed by an unsupervised machine learning algorithm that spread the data according to their most relevant characteristics. Results: The use of exercise data resulted in a better separation of healthy controls and HFPEF subjects than using rest data (spec=86.3% vs. spec=81.8% while the sensitivity does not substantially change). Notably, the stress protocol identified critical breathless subjects, which during exercise get close to the HFPEF subgroup while they were not differentiated from controls at rest (p-value in the figure). Conclusion: The characterization of HFPEF is significantly improved by analyzing velocities acquired during exercise. The method has strong clinical potential to identify critical breathless subjects at an early stage.