Abstract： The transient temperature fields and corresponding thermal deformations of a computer numerical control (CNC) turning center headstock were simulated using the finite element method. Based on the concept of thermal error sensitivity, the sensitivity images of thermal errors of the turning center headstock were derived from the simulation results. 512 candidate temperature key points were extracted from finite element nodes by using the wavelet image compression technique. According to the principle of the genetic optimization algorithm, the optimization object function was expressed as a flowchart which could be programmed as an Mfile in Matlab. The selection of temperature key points and building matching thermal error models were simultaneously realized. The validation experimental results on the turning center show that the fitting precision of thermal error models built on optimized temperature key points is better than that on arbitrarily selected temperature measuring points.