In Southeast China, the presence of boulders significantly threatens subway construction projects. Undetected boulders could negatively impact the efficiency and safety of subway construction methods that use shield and tunnel-boring machines. Therefore, boulder detection is a necessary step before construction begins. Recently, the cross-hole electrical resistivity tomography (ERT) method has gained attention as a potential boulder-detection approach in subway construction projects. However, in this method, the detection effect of each electrode configuration still needs to be assessed. In this study, a full-scale physical model test was conducted to determine the optimal electrode configuration for the cross-hole ERT method. Additionally, a sensitivity analysis of three types of electrode configurations, namely, bipole–bipole, dipole–dipole, and pole–tripole, was conducted. The results showed that the detection ability of the bipole–bipole electrode configuration was satisfactory, with very good boulder-detection resolution and sensitivity. Conversely, the resolution of the dipole–dipole electrode configuration, with relatively low signal strength, decreased as boulder distribution became more complex. Although the pole–tripole electrode configuration showed acceptable resolution, the boulder-detection results had many artifacts. Thus, the full-scale physical model test was conducted to thoroughly investigate the cross-hole ERT electrode configurations and determine the influence of boulder distribution on the measurements, with a view on achieving a more efficient application of this method in the field.