Abstract: The accurate evaluation of gas content is significant for the efficient exploration of unconventional natural gas reservoir, direct method adopts lost gas model and combines with desorption curve to evaluate reservoir gas content. However, the classical lost gas model is derived from constant pressure condition and spherical particle assumption in the estimation of coal bed methane, such as USBM method proposed by US Bureau of Mine, it brings a lot of errors for the deep-depth shale gas reservoir in which core sample is cylindrical shape. Based on diffusion theory, this work used time-varying pressure condition and cylindrical coordinate to solve one-dimensional diffusion equation and obtained analytical solution, then proposed a novel lost gas model (segmented variable boundary model). This model is enabled to describe two processes, i.e., drifting process and desorption process, with different gas-diffusion features. The result of segmented variable boundary model shows that the pressure drop between core sample and boundary increases in the drifting process when core sample is drifted from bottom hole to ground, due to the decreases of boundary pressure, lead to that the diffusion rate is accelerating when gas is escaping from core sample to environment, thus the diffusion curve of drifting process is concave. In the desorption process, core sample is placed in desorption canister and boundary pressure is constant, pressure drop between core sample and boundary is decreasing along the gas escaping from core sample, the diffusion rate is moderative and thus the curve of desorption process is convex. For further validating this model, a lost gas-desorption gas simulating experiment system was set up in lab based on the principle of similitude, and we conducted simulating experiment using cylindrical shale samples to obtain the diffusion curve in drifting process and desorption process, through comparing the experimental data with segmented variable boundary model and USBM model, demonstrated the validation of segmented variable boundary model. Moreover, the segmented variable boundary model is applied to fit the experiment data from the Y151 well in South Sichuan Basin, the fitting results have good consistency with experimental data, which indicates that the segmented variable boundary model is suitable for practical engineering condition.