SummaryThe most popular use of the Pura Raza Español horse in sport is for dressage competitions. Tests on young sport horses were first established in 2004 in Spain to collect data for the genetic evaluation of this breed's suitability for dressage. The aim of this study was to compare eight different models to find out the most appropriate way to include the rider in the genetic evaluation of dressage. A progressive removal of systematic effects from model was also analysed. A total of 8867 performance records collected between 2004 and 2011 from 1234 horses aged between 4 and 6 years old were used. The final score in the dressage test was used as the performance trait. The pedigree matrix contained 8487 individuals. A BLUP animal model was applied using a Bayesian approach with TM software. The horse's age, gender, travelling time, training level, stud of birth and event were included as systematic effects in all the models. Apart from the animal and residual effects that were present in all models, different models were compared combining random effects such as the rider, match (i.e. rider–horse interaction) and permanent environmental effects. A cross‐validation approach was used to evaluate the models' prediction ability. The best model included the permanent environmental, rider and match random effects. As far as systematic effects are concerned, the event or the stud of birth was essential effects needed to fit the data.