Machine tool thermal errors are an important element in machined workpiece inaccuracies. In the past few decades, thermal errors have been successfully reduced by software compensation techniques such as multiple linear regression analysis, finite element method, neural network and transfer function (TF). A phenomenological approach based on TFs is used for thermal error compensation in this research. This approach respects basic heat transfer mechanisms in the MT structure and requires a minimum of additional gauges. Every MT is unique due to manufacturing inaccuracies in components, assembly processes (different preloads in bolts, deviations in the assembly and seating of components, etc.) or working environment (air-conditioned room or ordinary production hall). The aim of this paper is to investigate thermal error compensation model transfer between machines of the same product line to improve thermal error model prediction. Several milling centers of the same product line were tested with varying non-stationary activities of heat sources from linear axis movements and in varying working environments. The issue of key temperature point selection is considered.