The bainite start temperature B s is defined as the highest temperature at which ferrite can transform by a displacive transformation. A common observation is that the bainite start temperature is very sensitive to the chemical composition, indicating that the influence of solutes is more than just thermodynamic. Empirical linear regression models have long been used to calculate the B s in a limited range of compositions. This paper attempts to create an empirical model of wider applicability and higher accuracy by means of neural networks. The results are compared with those calculated using the thermodynamic theory for bainite transformation, revealing that in general this theory agrees with the experimental results, but some discrepancies can still be found when the alloys are heavily alloyed.