The high-pressure mercury intrusion (HPMI) experiment is widely used to assess the pore architecture of tight sandstone reservoirs. However, the conventional analysis of the high-pressure mercury intrusion has always focused on the mercury injection curves themselves, neglecting the important geological information conveyed by the mercury ejection curves. This paper quantitatively describes the fractal characteristics of ejection curves by using four fractal models, i.e., Menger model, Thermodynamic model, Sierpinski model, and multi-fractal model. In comparison with mercury injection curves, we explore the fractal significance of mercury ejection curves and define the applicability of different fractal models in characterizing pore architectures. Investigated tight sandstone samples can be divided into four types (Types A, B, C and D) based on porosity, permeability, and mercury removal efficiency. Type D samples are unique in that they have higher permeability (>0.6 mD) but lower mercury removal efficiency (<35%). Fractal studies of the mercury injection curve show that it mainly reflects the pore throat characteristics, while the mercury ejection curve serves to reveal the pore features, and porosity and permeability correlate well with the fractal dimension of the injection curve, while mercury removal efficiency correlates only with the Ds' value of the ejection curve. The studies on the mercury ejection curves also reveal that the small pores and micropores of the Type C and Type D samples are more developed, with varying pore architecture. The fractal dimension DS' value of Type D samples is greater than that of Type C samples, and the dissolution of Type D samples is more intense than that of Type C samples, which further indicates that the Type D samples are smaller in pore size, rougher in surface, and with greater difficulty for the hydrocarbon to enter, resulting in their reservoir capacity probably less than that of Type C samples. In this regard, the important information characterized by the mercury ejection curve should be considered in evaluating the tight sandstone reservoirs. Finally, the Menger and Thermodynamic models prove to be more suitable for describing the total pore architecture, while the Sierpinski model is better for characterizing the variability of the interconnected pores. [ABSTRACT FROM AUTHOR]