Discus throwing performance is associated with technique variables. However, the relationships between technique variability and performance in discus throwing are unknown. Video clips of three throws each by 18 male and 15 female discus throwers were collected during a competition. Two-dimensional coordinate data were manually digitised and transferred into three-dimensional coordinates of body landmarks. Eight joint angles were calculated and their values were time-normalised to 101 frames from the initiation to the end of a throw. The mean standard deviations of 101 frames for eight joint angles among the three throws were calculated as the variability measurement. The best official distance of three throws was used as the performance measurement. For male throwers, the canonical correlations between technique variability and performance were significant (P < 0.1). Pearson correlations showed that the performance was significantly negatively correlated with the mean standard deviations of arm-shoulder separation angle (r = -0.57, P = 0.02), hip-shoulder separation angle (r = -0.63, P < 0.01), left hip flexion angle (r = -0.5, P = 0.03), and trunk forward-backward tilt angle (r = -0.57, P = 0.02). For female throwers, the canonical correlations between technique variability and performance were not significant (P > 0.1). Male discus throwers with a longer throwing distance had smaller variability in reproducing techniques. Reducing motion variability may be an important goal for discus training in males.