This study compares the goodness of fit of 10 empirical models used to predict gas diffusivity (Dp D0−1) with experimental data obtained from Chilean soil samples from different soil management practices. Nine sites under different soil management practices were sampled at different depths. In total, 275 soil cores were obtained. The gas diffusion coefficient (Dp) was determined at different matric potentials using a gas diffusion chamber saturated with free-oxygen (O2) nitrogen (N2) as the gas that diffuses and oxygen as the measured gas inside the diffusion chamber with a gaseous oxygen sensor. Complementary soil properties were measured in order to modelate the diffusivity with several models. The use of statistical indexes, i.e., determination coefficient (r2), root mean square error (RMSE), mean bias error (BIAS), the agreement index (d), and the mean absolute error (MAE), to rank the models according to the fit of goodness was proposed. The models of Millington and Quirk (M-Q), Penman Water Linear Reduction Model (P-WLR), Millington Water Linear Reduction Model (MI-WLR), and Marshal Water Linear Reduction Model (MA-WLR) showed a high simplicity and had a better prediction of gas diffusivity than more complex models. The Three-Porosity Model (TPM) showed the worst performance among the models. Thus, the use of more complex models does not guarantee a better prediction of gas diffusivity. However, it is necessary to test other complex models that incorporate soil management practices and have presented better results than those used in this work. Also, incorporating new soil management could be the base to develop a more accurate comparison. Finally, the P-WLR and TPM models had the best and worst performances above all models. It is suggested to test new models and to increase soil management in future research.