<p>Spindle speed regulation is a common way to stabilize chatter and improve the material removal rate during machining. The corresponding surface location error (SLE) prediction method is necessary for the selection of the appropriate spindle speed that satisfies accuracy requirements. However, the identified errors of model parameters and the variations in the mechanics and dynamics of the machining process caused by changing spindle speeds result in inaccuracies in SLE prediction, which could not be reduced only based on the mechanistic model. This paper presents a novel method combining the mechanistic model with a data model for SLE prediction. Firstly, the actual SLE is decomposed into two portions, one of which can be predicted based on the mechanistic model and the other is caused by many uncertain factors. Then, the mechanistic model is developed depending on the dynamics model of the machining system and the spindle speed-based cutting force model. Based on Bayesian inference and a few measured data, the mechanistic model is updated to reduce the prediction errors caused by uncertain factors. The developing procedure of the mixture model combining the mechanistic and data models is proposed, and the updated value and domain of SLE prediction are obtained. The proposed method is validated based on the milling test with varying spindle speeds and the SLE measurements on a benchmark with ten thin plates. It is shown that the average prediction error of the proposed method for varying spindle speed SLE is improved by 42% compared to the mechanistic-only method. Meanwhile, the domain of SLE distribution in varying spindle speeds can be accurately predicted.</p>