<p>Total organic carbon (TOC) content is a crucial evaluation parameter in the process of shale gas exploration and development. Marine-continental transitional shale is characterized by strong heterogeneity and thin single-layer thickness. The discrete TOC data measured by experimental methods are unable to accurately reflect the reservoir characteristics of marine-continental transitional shale. In this paper, a multivariate nonlinear regression prediction model (R-MNR) was established, and the model was applied to predict the TOC content of shale for the first time. The ΔlgR model, multiple linear regression model (MLR), BP neural network model (BP model), and R-MNR model were built to predict the TOC of shale in Benxi Formation. The coefficient of determination (<i>R</i>²), mean-absolute-percentage-error (MAPE), root-mean-square-error (RMSE), and the number of input layer parameters (NILP) were employed to assess the efficacy of the model through the analytic hierarchy process (AHP) method. The total weight of R-MNR is 0.361, and that of BP model is 0.336. The weights of the two traditional models are 0.104 and 0.199, respectively. The results indicate that the R-MNR is comparable to the BP model in terms of prediction accuracy, and both models are significantly more accurate than the traditional prediction model. The R-MNR is capable of obtaining a clear TOC prediction formula, which is convenient for verification and promotion. During the training process of the R-MNR, the influence of each parameter and coupling relationship on the prediction results is elucidated, which enables researchers to gain a deeper understanding of the geophysical significance and geological process of the model. The result of this study suggests that the R-MNR can be employed to predict the TOC content of marine-continental transitional shale effectively in the future.</p>