Abstract Heart failure with preserved ejection fraction (HFpEF) is a highly heterogeneous syndrome associated with multiple medical comorbidities and pathophysiologic pathways or phenotypes. We recently developed a phenomapping method combining deep phenotyping with machine learning analysis to classify HFpEF patients into 3 clinically distinct phenotypic subgroups (pheno-groups) with different clinical outcomes. Pheno-group #1 was younger with lower B-type natriuretic peptide (BNP) levels, pheno-group #2 had the highest prevalence of obesity and diabetes mellitus, and pheno-group #3 was the oldest with the most factors for chronic kidney disease, the most dysfunctional myocardial mechanics, and the highest adverse outcomes. The pathophysiological differences between these pheno-groups, however, remain incompletely described. We sought to evaluate whether these 3 groups differ on the basis of repolarization heterogeneity, which has previously been linked to adverse outcomes in HFpEF. The T-peak to T-end (TpTe) interval, a well-validated index of repolarization heterogeneity, was measured by 2 readers blinded to each other and all other clinical data on the electrocardiograms of 201 HFpEF patients enrolled in a systematic observational study. TpTe duration was associated with higher BNP level (P=0.006), increased QRS-T angle (P=0.008), and lower septal e’ velocity (P=0.007). TpTe duration was greatest in pheno-group #3 (100.4±24.5 ms) compared to pheno-groups #1 (91.2±17.3 ms) and #2 (90.2±17.0 ms) ( P= 0.0098). On multivariable analyses, increased TpTe was independently associated with the high-risk pheno-group #3 classification. In conclusion, repolarization heterogeneity is a marker of a specific subset of HFpEF patients identified using unsupervised machine learning analysis and therefore may be a key pathophysiologic marker in this subset of HFpEF patients.