To quantitatively evaluate how setup errors in conjunction with dose gradients contribute to the error in IMRT dose quality assurance (DQA) measurements. The control group consisted of 5 DQA plans of which all individual field dose differences were less than ${pm}5%$. On the contrary, the examination group was composed of 16 DQA plans where any individual field dose difference was larger than ${pm}10%$ even though their total dose differences were less than ${pm}5%$. The difference in 3D dose gradients between the two groups was estimated in a cube of $6{ imes}6{ imes}6;mm^3$ centered at the verification point. Under the assumption that setup errors existed during the DQA measurements of the examination group, a three dimensional offset point inside the cube was sought out, where the individual field dose difference was minimized. The average dose gradients of the control group along the x, y, and z axes were 0.21, 0.20, and 0.15 $cGy{cdot}mm^{-1}$, respectively, while those of the examination group were 0.64, 0.48, and 0.28 $cGy{cdot}mm^{-1}$, respectively. All 16 plans of the examination group had their own 3D offset points in the cube. The individual field dose differences recalculated at the offset points were mostly diminished and thus the average values of total and individual field dose differences were reduced from 3.1% to 2.2% and 15.4% to 2.2%, respectively. The offset distribution turned out to be random in the 3D coordinate. This study provided the quantitative data that support the large individual field dose difference mainly stems from possible geometric errors (e.g., random setup errors) under the influence of steep dose gradients of IMRT field.