To evaluate the accuracy of compressed sensing (CS) cardiovascular magnetic resonance imaging (CMR) in the classifications of heart failure (HF). One hundred and fifty-four patients underwent 3 T CMR using CS single-breath-hold cine (SBH-cine) after a standard multiple-breath-hold cine (MBH-cine). The image quality of the two methods was compared. Cardiac function parameters were quantitatively analyzed. The patients were divided into different HF types based on diagnostic criteria using two sequences. The scan time for CS SBH-cine was reduced by 88% compared with that of standard MBH-cine. In the non-atrial-fibrillation (NAF) group (n = 121), the image quality score of CS SBH-cine was slightly decreased compared with standard MBH-cine (4.5 ± 0.6 for the CS SBH-cine vs. 4.7 ± 0.5 for the standard MBH-cine, T = 5.038, p < 0.05). In the atrial fibrillation (AF) group (n = 33), the CS SBH-cine image quality score was slightly higher than that of the standard MBH-cine (3.8 ± 0.7 for the CS SBH-cine vs. 3.3 ± 0.5 for the standard MBH-cine, T =  - 4.503, p < 0.05). The left ventricular (LV) end diastolic volume, LV end systolic volume, LV ejection fraction, and LV mass index calculated by CS SBH-cine had no significant differences from those calculated by standard MBH-cine. The agreement on HF classifications of both the standard MBH-cine and the CS SBH-cine was excellent (kappa = 0.934, p < 0.05). The result showed that HF could be classified accurately using CS SBH-cine.