Considering　that　some　linear　and　nonlinear　factors　to　thermal　error　data　exist　when　a　machine　tool　works,　this　paper　proposes　a　modeling　method　for　prediction　of　machine　tools＇　thermal　errors　by　using　a　grey　linear　regression　combination　thermal　error　model.　This　method　has　an　ability　to　deal　with　the　linear　and　nonlinear　problems.　To　obtain　predictive　values　of　thermal　errors,　its　residual　error　is　corrected　by　the　BP　neural　network.　The　predictive　value　obtained　from　a　grey　model　using　an　exponential　function　to　simulate　the　data,　is　compared　with　the　one　obtained　above,　and　the　result　proves　the　superiority　of　the　grey　linear　regression　combination　and　the　BP　neural　network　model　for　machine　tools＇　thermal　error　compensation　modeling.