The thermal positioning error of linear axis is related to position and temperature at the same time, and the traditional modelling method has a heavy workload, low efficiency and poor prediction accuracy under variable condition. To address this issue, a decoupling and step-by-step modelling method for thermal positioning error of linear axis is proposed in this article. Firstly, the thermal positioning error measured is decoupled, and the slope parameters and intercept parameters only related to the temperature are obtained based on least square linear fitting. Secondly, the absolute temperature and relative temperature are used to build the slope parameter model and intercept parameter model step by step, and the mapping relationship is obtained. Combined with the slope and intercept, the thermal positioning error model is formulated. Finally, based on the established model, the thermal positioning error in a new working condition is predicted, and the maximum residual error can be realized as 1. 6 μm. Compared with the direct modelling method, the prediction accuracy of the proposed method is improved greatly, which shows its effectiveness.