Arrhythmias are commonly reported in exercising horses, however due to regulatory constraints electrocardiograms (ECGs) are acquired during training but not competition, raising questions about the repeatability of findings. The aims were (1) compare training and competition arrhythmias and (2) describe the repeatability of arrhythmias during maximal-intensity exercise. A convenience sample of 52 healthy Thoroughbreds (aged 8.7 ± 2.5 years) competing in the World Professional Chuckwagon Association were obtained, totaling 152 training or competition ECGs (2–7 ECGs/horse). Speed, heart rate (HR) and arrhythmias (supraventricular premature complex, SVPC; ventricular premature complex, VPC) were examined. Pre- and post-recovery (approximately 6 min) blood samples measured lactate and high-sensitivity troponin-T. Training and competition arrythmias were compared (Friedman’s test) and reliability of repeated ECGs assessed (intraclass correlation; P < 0.05). Training vs. competition: Forty horses had clean tracing from training and competition (n = 80 ECGs); the number and type of arrhythmias were not different. In training, VPCs were present in 7/40 horses (median [interquartile range, IQR]/ECG; range; 0 [0,0]; 0–4) and 9/40 horses (0 [0,0]; 0–5) in active-recovery. In competition, VPCs were present in 7/40 horses (0 [0,0]; 0–8) and 8/40 horses (0 [0,0]; 0–5) in active-recovery. Arrhythmias were primarily single premature complexes. Training and competition speed, HR, lactate and troponin-T did not differ however, sampling was too early for peak serum Troponin-T levels. Repeatability: total arrhythmias between serial ECGs did not differ. The reliability to detect SVPCs and VPCs was poor to moderate, and poor, respectively. Overall, the total number of arrhythmias was repeatable, but the reliability of arrhythmia type was poor to moderate.