Cocoa bean roasting allows for reactions to occur between the characteristic aroma and taste precursors that are involved in the sensory perception of chocolate and cocoa by-products. This work evaluates the moisture kinetics of cocoa beans during the roasting process by applying empirical and semi-empirical exponential models. Four roasting temperatures (100, 140, 180, and 220 °C) were used in a cylindrically designed toaster. Three reaction kinetics were tested (pseudo zero order, pseudo first order, and second order), along with 10 exponential models (Newton, Page, Henderson and Pabis, Logarithmic, Two-Term, Midilli, Verma, Diffusion Approximation, Silva, and Peleg). The Fick equation was applied to estimate the diffusion coefficients. The dependence on the activation energy for the moisture diffusion process was described by the Arrhenius equation. The kinetic parameters and exponential models were estimated by non-linear regression. The models with better reproducibility were the pseudo first order, the Page, and the Verma models (R2 ≥ 0.98). The diffusion coefficients that were calculated were in the order of 1.26 to 5.70 × 109 m s−2 and the energy activation for moisture diffusion obtained was 19.52 kJ mol−1.