Radiative Transfer Models (RTMs) are one of the major building blocks of remote sensing data analysis that are widely used for various tasks, such as atmospheric correction of satellite imagery. Although high-fidelity physical RTMs like MODTRAN offer the best possible modeling of atmospheric procedures, they are computationally demanding and require many hyperparameters that are difficult to set by a nonprofessional user. Therefore, there is a need for surrogate models for the physical radiative transfer codes that can mitigate these drawbacks while offering an acceptable performance. This study aimed to suggest surrogate models for the MODTRAN RTM using machine learning and deep learning algorithms. For this purpose, the top-of-atmosphere (TOP) spectra were calculated by the MODTRAN code, and the bottom-of-atmosphere (BOA) input spectra and other atmospheric parameters like temperature and water vapor content observations were collected for the training dataset. Three deep learning regression models, including a fully connected network (FCN), a 1-D convolutional neural network (CNN), and an auto-encoder (AE), as well as the random forest (RF) machine learning regression model, were trained using the collected dataset. The results of these models were assessed using three evaluation metrics of root mean squared error (RMSE), regression coefficient (R²), and spectral angle (SAM). The evaluations indicated that the AE offered the best performance in all the metrics with RMSE, R², and SAM scores of 0.0047, 0.9906, and 1.3987 (degrees), respectively, in the best-case scenarios. Moreover, the random forest model performed worst with RMSE, R², and SAM scores of 0.0077, 0.9507, and 2.1443 (degrees) in the best-case scenarios. These results proved the highly non-linear nature of the radiative transfer codes and showed that the deep learning models could better model the high-fidelity physical RTMs.