This paper presents the kinematic calibration of a four-degrees-of-freedom (4DOF) high-speed parallel robot. In order to improve the calibration effect by decreasing the influence of the unobservable disturbance variables introduced by error measurement, a measurement configuration optimization method is proposed. Configurations are iteratively selected inside the workspace by a searching algorithm, then the selection results are evaluated through an index associated with the condition number of the identification Jacobian matrix; finally, the number of optimized configurations is determined. Since the selection algorithm has been shown to be sensitive to local minima, a meta-heuristic method has been applied to decrease this sensibility. To verify the effectiveness of the algorithm and kinematic calibration, computation validations, pose error estimations, and experiments are performed. The results show that the identification accuracy and calibration effect can be significantly improved by using the optimized configurations.