Twinning is a serious problem in Thoroughbred horses, as most of the multiple pregnancies are miscarried and live foals present lower performance values. The aim of this study was to estimate the heritability and repeatability of twinning in Thoroughbred horses. A total of 12648 pregnancies of 2033 Thoroughbred mares born between 1929 and 1994 in 11 studs were included. The analysis was performed by the use of AI-REML algorithm under three threshold animal models. The first model included effects of stud, mare's birth period and interaction between them as fixed. The second model included the interaction between the stud and mare's birth period as random, whereas in the third model stud effects were also treated as random. The models were compared based on the deviances. Phenotypic and genetic trends were evaluated as well. The twinning rate was 3.5%. The study involved 443 twins and two triplets. Heritability estimates varied from 0.24 (model III) to 0.29 (model I), whereas repeatability estimates ranged from 0.33 (model I) to 0.34 (model III). The fluctuations of average annual genetic effects may result from segregation of the major gene in the population.