The counter-flow hollow fiber membrane-based direct evaporative cooler (MDEC) was proposed as an energy-efficient and hygienic air cooling solution. The random sequential addition algorithm was used to simulate the randomness of fiber distribution within the MDEC in engineering practice. 3D numerical models were developed with a computational domain determined as a hexagonal prism containing multiple fibers. The experimentally validated model depicted the air state variation along the path under regular and random fiber distributions. The differences in wet-bulb effectiveness (εwb) and coefficient of performance (COP) of MDEC with two configurations, regular and random fiber distributions, were compared for different packing fractions and air flow rates. Furthermore, dimensionless correlations for the shell-side friction factor (f), Nusselt number (Nu), and Sherwood number (Sh) were derived to generalize the results to other hollow fiber membrane modules employed in liquid/gas systems. The results showed that the channel flow effect and dead zone under the random configuration worsened the heat and moisture handling performance of MDEC. Compared with the regular configuration, both Nu and Sh were degraded by 11.6–55.6%, but f was desirably reduced by 28.8–49.8%. Balancing cooling capability and energy efficiency, a regular configuration with moderate packing fraction was more beneficial for engineering practice.