This paper presents a thermal error model for a lathe CNC machine using temperature on moving parts, which is usually a difficult measurement. Different working conditions are performed to obtain temperature and thermal displacement of the machine. A multi-linear regression model is applied for modelling relationship between temperatures and thermal errors. Data of two constant working conditions are implemented to find out the fitting function. Through analysis, it shows that thermal errors for X- and Z-axes can be estimated based on only four temperature points (T 2 , T 3 , T 10 and T 12 for the X-axis and T 4 , T 6 , T 7 and T 11 for the Z-axis). Results reveal that using movable part temperature data (T 12 ), the average and maximum error in the predicted thermal displacement of the X-axis can be reduced about 1/3 and 2/5 compared to without using T 12 , while the movable temperature point on Z-Screw (T 8 ) has a tiny effect on the average and maximum error of predicted thermal displacement of the Z-axis. Results herein can provide useful information to develop a thermal compensation for the CNC lathe machine.