The physical properties of tight shale reservoirs have always been a hot topic of discussion, and the tuff reservoirs have also attracted more and more attention. Log predicting of the porosity is also important for the exploration and development of tight oil. Tight oil resources have been found in tight shale and tuff reservoirs in the Chang 7 Member of the Upper Triassic Yanchang Formation in the Ordos Basin, China. However, the logging method that effectively predicting porosity of the Chang 7 Member is lacking, and few people have discussed how to optimize the log interpretation methods. In this study, the characteristics of shale and tuff reservoirs were summarized by microscope and scanning electron microscope observation, X-ray diffraction, pore and permeability experiment, and the logging data. In order to find the optimal method of porosity prediction, four methods were performed: multiple regression fitting, multi-component volume model, porosity logging formula, and back propagation neural network method.The “Zhangjiatan Shale” of the Chang 7 Member is characterized by abundant sandy lamina and tuff lamina, compared with the tuff reservoir, the enrichment of pyrite (average 22.7 wt%) and total organic matter (>6%) increase the content of intercrystalline pores and organic matter related pores, the logging response is characterized by high natural gamma ray, low spontaneous potential and high acoustic. The tuff reservoir is mainly composed of vitric tuff and crystal-vitric tuff, hence, the rich in intergranular pore and micro-fracture can be explained by the dissolution and devitrification of vitric and crystal fragments, and the logging characteristic is opposite to that of the “Zhangjiatan Shale”. Four methods for predicting porosity are established. First, appropriate logging parameters are selected for regression fitting in the “Zhangjiatan Shale” (e.g., gamma ray, deep lateral logging resistivity, density and spontaneous potential) and tuff reservoir (e.g., acoustic, density, compensated neutron logging). Then, the rock multi-component volume model was established as rock skeleton, clay or tuff matter, organic matter, and pores. The logging parameters of the skeleton, clay and organic matter were obtained by intersection graphs, which were applied in the improved porosity logging formulas. The effect of organic matter on logging was considered into the traditional porosity prediction formula, a third prediction method was established. Finally, the BP method was introduced into the porosity prediction work, and the neural network parameters were adjusted according to the log and porosity data. The learning rate and the learning accuracy were set as 0.1 and 10−6 in the study area. The error analysis between the result predicted and measured shows that the multi-component volume model is suitable for the “Zhangjiatan Shale” (the correlative coefficient R>80%), while the back propagation neural network is the best method for the tuff reservoir (the correlative coefficient R>90%).