Steel moment frames are a popular seismic-force resisting system, but it is believed that they are susceptible to early fracture if there is a stress concentration in the plastic hinge region, also known as the protected zone. If a defect is present in this area, it may be repaired by grinding and/or welding, but little research has investigated how the repairs affect the performance of full-scale moment connections subjected to inelastic rotations. Thus, the goals of this research were to establish the performance of full-scale moment connections with repairs and defects, then develop a method for predicting fracture of the full-scale specimens using more economical cyclic bend tests. To do this, six full-scale reduced beam section (RBS) connections were tested having arrays of repairs or defects applied to the flanges. The repairs were 0.125 in. deep notches ground to a smooth taper and 0.25 in. deep notches ground to a smooth taper, welded, and ground smooth. The defects were sharp 0.25 in. and 0.375 in. notches. In addition, 54 bend tests were conducted on beam flange and bar stock coupons having the same repairs and defects, power actuated fasteners, puddle welds, and no artifacts. Finally, Coffin-Manson low-cycle fatigue relationships were calibrated using results from the cyclic bend tests with each artifact (repair, defect, or attachment method) and used in conjunction with estimates of full-scale plastic strain amplitudes to predict fracture of full-scale specimens. All four of the full-scale moment connections with repairs satisfied special moment frame qualification criteria (SMF). One full-scale specimen with sharp 0.25 in. notches satisfied SMF qualification criteria, but the flexural resistance dropped rapidly after the qualification cycle. On the other hand, the specimen with sharp 0.375 in. notches did not satisfy SMF qualification criteria due to ductile fractures propagating from the notches. The proposed method for predicting fracture of full-scale connections was validated using the six current and six previous full-scale RBS specimens. This method underpredicted fracture for eleven of the twelve specimens. The ratio of the actual to predicted cumulative story drift at fracture had a mean of 1.13 and a standard deviation of 0.19.