Discrete fracture network (DFN) modeling is widely applied to characterize fracture systems in tight reservoirs, yet its accuracy is often limited by the insufficient integration of geological, geometric, and mechanical controls. To address these challenges, a fracture network modeling approach based on mechanical-geometric criteria (FNMmg) is introduced. The method incorporates five analytical strategies: (1) simulation of stratabound fracture dips using gap angles measured relative to bedding normals; (2) scale-weighted orientation statistics to emphasize the contribution of dominant fractures; (3) orientation generation via a cumulative probability curve to avoid errors caused by fracture-group partitioning; (4) fracture element representation to enable hierarchical coalescence; and (5) fusion rules combining geometric similarity with mechanical and diagenetic principles. Application to the Triassic Chang 6 tight sandstone reservoir in the Ordos Basin, China, demonstrates that FNMmg can reproduce multi-scale fracture systems with improved geological consistency. Validation against tracer monitoring data from 34 injector-producer well pairs showed an overall connectivity consistency of about 91%, including 86% accuracy for fracture-related and 93% for pore-related or unconnected cases, demonstrating FNMmg's utility in identifying effective fracture flow paths and supporting reservoir connectivity analysis. These results indicate that FNMmg provides a practical framework that explicitly integrates fracture mechanics, geometry, and data-driven constraints, offering a basis for reservoir characterization, connectivity analysis, and development optimization in tight sandstone settings.