Abstract: Due to the poor physical properties of tight reservoirs, CO2 huff-n-puff (HNP) is considered a potential enhanced oil recovery (EOR) method after primary depletion. Optimization plays a critical role in the effective implementation of CO2 huff-n-puff. But the optimization requires a good understanding of the EOR mechanisms. In this work, the spatial distribution of oil saturation under different experimental conditions was analyzed by the NMR method to further discuss the HNP mechanisms. According to the variation of 1D frequency signal amplitude, we divided the core into the hardly movable area and movable area, the region with the obvious signal decline was defined as the movable area, and the hardly movable area was the region with limited signal decline. Based on that the recovery characteristics of different scenarios were evaluated. Firstly, the necessity of the soaking stage was studied, where three scenarios with different soaking times were carried out. Secondly, the injection pressure was adjusted to investigate the effect of the pressure gradient. The T2 spectra show that soaking has significantly improved the production of crude oil in small pores, and higher oil recovery in a single cycle is observed, but it is lower when the elapsed time (total operation time) is the same. 31.03% of oil can be recovered after 3 cycles HNP, which increases to 33.8% and 37.06% for the 4 cycles and 6 cycles cases. As the pressure gradient increases, more oil is removed out of the matrix, and the oil in the deep part of the reservoir can be effectively recovered. During the CO2 huff-n-puff process, the oil distributions are similar to the solution gas drive, the residual oil is distributed at the close end of the core and the range that the oil can be efficiently recovered is limited.