The main objective of this paper is to develop predictive models using Beta regression for laboratory-prepared hot mix asphalt (HMA) specimens' thermal properties, including thermal conductivity (TC), thermal diffusivity (TD) and specific heat (SH). Thirty such specimens were prepared while varying the mixture's nominal maximum aggregate sizes (NMAS) and gradation coarseness. The widely used Transient Plane Source (TPS) method was employed to determine the thermal properties of the asphalt concrete. Only one type of asphalt binder was used for preparing all specimens. The air void volume (Va) and the effective binder volume (Vbe) were calculated for each mixture. To this end, the multiple linear regressions and the non-linear beta regressions were employed. Laboratory work resulted in hundred and fifty (150) data points. Three nominal maximum aggregate sizes, two gradation coarseness levels, five replicates and five different locations of measurements to ensure accuracy and repeatability in the obtained results. In conclusion, using Va and Vbe as predictors provided reliable predictive models for the thermal properties of different asphalt mixtures. The distribution of Va and Vbe was identified, and synthetic data was created to evaluate the accuracy of the models. Apart from R2 values, beta regression was more reliable to predict thermal properties of asphalt mixtures than multiple linear regression.