Using special anisotropic membranes of bilayered cylindrical channels, we have experimentally clarified the origin of inconsistency in the transition widths and critical flow rates (q(c)) reported in related ultrafiltration studies of flexible linear chains. By studying the retention behavior of linear polystyrene chains passing through the same bilayered ultrafiltration membrane but in different translocation models, we reveal that the interaction among flow fields and the preconfinement effect are responsible for the previously observed inconsistency. In theory, only a single pore and a single chain are considered, but many pores and chains exist in experiment. The interaction among flow fields generated at different pore entrances could lead to the elongation/turbulent mixed flow fields, which will result in some unpredictable motions of polymer chains, such as the rotation, compression, reversed pulling, etc. Accordingly, much larger q(c) values and a broader transition were observed in experiments. On the basis of our previous free-draining model and present results, we propose a revised model by considering the partially draining nature of one confined blob, which provides a simple way for the rough estimation of the degree of draining for one confined blob in good solvents. Besides, the results also reveal that the preconfinement effect could lead to a reduced conformation entropy and an increased preconfinement energy for a polymer chain before its translocation through a small cylindrical pore, but such an influence seems to be chain length independent when the normalized confinement energy is compared. It satisfactorily explains why the critical flow rate is widely reported to be chain length independent, even though the preconfinement effect is not taken into consideration in related studies. Finally, our control experiment reconfirms that the asymmetry of bilayered membrane, instead of the variation of effective pore size, is the origin for the different ultrafiltration behavior of linear chains in different translocation models.