Despite strong selection for athletic traits in Thoroughbred horses, there is marked variation in speed and aptitude for racing performance within the breed. Using global positioning system monitoring during exercise training, we measured speed variables and temporal changes in speed with age to derive phenotypes for GWAS. The aim of the study was to test the hypothesis that genetic variation contributes to variation in end‐point physiological traits, in this case galloping speed measured during field exercise tests. Standardisation of field‐measured phenotypes was attempted by assessing horses exercised on the same gallop track and managed under similar conditions by a single trainer. PCA of six key speed indices captured 73.9% of the variation with principal component 1 (PC1). Verifying the utility of the phenotype, we observed that PC1 (median) in 2‐year‐old horses was significantly different among elite, non‐elite and unraced horses (P < 0.001) and the temporal change with age in PC1 varied among horses with different myostatin (MSTN) g.66493737C>T SNP genotypes. A GWAS for PC1 in 2‐year‐old horses (n = 122) identified four SNPs reaching the suggestive threshold for association (P < 4.80 × 10−5), defining a 1.09 Mb candidate region on ECA8 containing the myosin XVIIIB (MYO18B) gene. In a GWAS for temporal change in PC1 with age (n = 168), five SNPs reached the suggestive threshold for association and defined candidate regions on ECA2 and ECA11. Both regions contained genes that are significantly differentially expressed in equine skeletal muscle in response to acute exercise and training stimuli, including MYO18A. As MYO18A plays a regulatory role in the skeletal muscle response to exercise, the identified genomic variation proximal to the myosin family genes may be important for the regulation of the response to exercise and training.