The present work studies, using statistical methods, the features of austenite grain size distributions (AGSDs) obtained in an ASTM A213-T91 steel by applying two-step thermal cycles, namely, thermal treatment at tempering temperature +'in situ' austenitisation. Changes in AGSD were produced by varying the holding time at tempering temperature and the heating rate to austenite. The application of different thermal cycle conditions induced two main characteristics in the AGSDs, i.e. the growth of a few very large grains in a matrix of small to medium sized grains, that is, heterogeneous grain size, and a bimodal grain size distribution in that matrix. Statistical techniques were used to describe the resulting AGSDs and to investigate a possible correlation between bimodality and heterogeneity. The characterisations have been carried out using the kernel density estimation (KDE) technique to compute the underlying statistical distribution, and then applying a significance test able to assess whether or not bimodality, if detected with KDE, is statistically significant. Results indicate that bimodality is evidenced for the higher heating rates to austenite and is very likely related to the development of the heterogeneous grain size structure. © 2014 Institute of Materials, Minerals and Mining.