The progression of damage in gradient stress fields under cyclic loading was studied in simply-supported graphite/epoxy beam-columns. Three layups, [454/-454/(0/90)412s, [±45/0/904]4s, and [(452/-452/0)2/905]2s, were chosen to be consistent with data collected in a previous study with static loading. Specimens were loaded statically until predetermined characteristic damage levels were obtained, after which cyclic loading began. Cyclic tests were run in load control at an R ratio of ten and a frequency of 1.5 Hz. At specific intervals, cyclic loading was stopped, a static test measuring the load and the corresponding center deflection was performed, and edge replicas of the specimen's sides were taken. Cyclic loading was then resumed until the next interval or final failure. Damage histories were pieced together for each laminate type under different maximum cyclic load levels. The damage of these specimens varied both along the length and through the thickness of the specimens with two modes of damage present. Damage due to the static loading occurred on the tension side of the specimen and was controlled by matrix cracks. Delaminations initiated and grew in areas of crack saturation. Damage due to the cyclic loading also occurred on the compression side of the specimens and was controlled by delaminations which initiated and grew independently of matrix cracks, leading to sublaminate buckling. These compression side delaminations did not initiate in static tests to failure. Increasing the maximum cyclic load changed the relative growth of the two modes of damage, thus changing the failure mode. The stiffness of the specimens increased slightly with cyclic loading but steadily declined thereafter. Stiffness degradation was less than 6% until just prior to failure. Because cyclic loading revealed critical damage modes not found with static loading, cyclic loading should be used when ascertaining all possible damage modes. Using higher cyclic load levels to reduce testing time can hide critical damage modes with the potential of overestimating cyclic lifetime.