Genetic evaluations of sport performance typically consider competition records of ranking points in each competition, accumulated lifetime points or annual earnings. Repeated observations have the advantage of allowing for adjustment of effects associated with each competition such as rider experience, judge and competing horses, but also demands more computer capacity than single-trait records, which could prohibit multiple-trait evaluations. The aim of the study was to compare CPU times, estimated breeding values (EBVs), reliabilities and model prediction abilities when modelling repeated competition ranking points (run A), mean ranking points (runs B and C), mean ranking points precorrected for effects associated with each competition (run D) and accumulated lifetime points (run E) for Danish Warmblood horses. CPU times for run A were 632–776 times (show jumping) and 59–96 times (dressage) as high as for runs B–E. EBVs of run D were perfectly correlated (1.00) with those of run A. Reliabilities were highest in runs E and A. Best model prediction ability and least bias were found in run C (dressage) and run E (show jumping), but the best choice in each discipline was not preferable for the other. Run D was the second best in both disciplines (D), and is expected to increase in performance over time as omission of a relatively large amount of historic data becomes less important.