The dependency of the martensite start (Ms) temperature upon composition of engineering steels has been examined by analyzing the results predicted by an artificial neural network (ANN) model and thermodynamic data. Two new formulas, the simple linear and binary interaction ones, have been statistically derived and applied to predict the Ms temperature in an Fe–C–Si–Mn–Cr–Mo system. It is shown that the separation of the influence of interactions from that of individual alloying elements is successful since most of the statistical results are reasonable and thus have been physically interpreted. The thermodynamic calculations show that the alloying elements have similar influence upon the Ms and A₃ temperatures. The apparent effect of carbon depends largely on C–X interactions. C–Mn and C–Mo interactions weaken the effect of carbon while that of C–Si interaction intensifies the role of C. This is supported by phenomenological results and has been physically interpreted. The interactions between substitutional alloying elements have also significant influence upon the Ms temperature. The Si–Mn interaction strongly increases the Ms, while Si–Mo interaction significantly decreases the Ms. So far, there is no proper physical explanation for this though supportive evidence has been obtained from phenomenological results. Mn and Mo have the weakest apparent interaction, that is, their influence can be simply added up. Moreover, a semi-physical model has been built to predict the Ms temperature from a critical temperature, which can be calculated thermodynamically. It shows that the semi-physical method gives a satisfactory prediction of Ms with a standard error of 15.3°C . Evaluation of nine common empirical methods indicates that the Kung and Rayment (KR) formula gives the best predicting results amongst them.