Total Organic Carbon content (TOC) is one of the key indicators for shale gas resource evaluation and prediction. Prediction from well logging data is an important mean to reveal the continuous variations TOC. This work aims to uncover the effects of all kinds of forecasting methods in upper Paleozoic Marine shale, Methods: Different prediction methods (Δ LogR method and its improved method, multiple linear regression method and neural network) in different research scales (segment or layered) were employed to establish TOC logging prediction model of the first member of Longmaxi Formation, Changning area, Southern Sichuan Basin. Results: The results show that the prediction effect and applicability of each method are all different. In general, the prediction effects of multiple linear regression method and BP neural network method are much better than ΔLog R method and its improved method in study area. In this work, the multiple linear regression method exhibits better effect in predicting the high TOC segment, while the neural network method display more accurate effect in predicting low TOC value segment. Conclusions: The method of "fine stratification and optimal method matching" is proposed to select the corresponding method for prediction according to the distribution of organic matter and logging response characteristics of different layers in Long1. For micro-layer a, b, c of Lower Long1 subsection (Long11a-c) and micro-layer d of Lower Long1 (Long11d) -Upper Long1 subsection (Long12) , multiple linear regression method and BP neural network method were used for modeling, and the best prediction effect was obtained. Not only the prediction accuracy was high, but also the relative error was small, the error of most samples was less than 20%.