Deep inspiration breath hold (DIBH) is often used to immobilize pancreatic tumor for SBRT. Geometric reproducibility of tumor positioning between breath hold have been reported, but the corresponding dosimetric impact is not well understood. We evaluated dosimetric reproducibility of tumor coverage and OAR doses for inter-breath hold variation at treatment. Ten pancreatic cancer patients who underwent DIBH SBRT with 660cGy x 5 fractions (Rx) were selected for the study. The implanted fiducial markers were used as tumor motion surrogate at treatment. At CT sim, the contrast planning CT was acquired with 2-mm slice thickness, immediately followed by three additional CT sets. The patient specific breath-hold variation measured from the four sim CT sets was added as a margin to GTV in addition to a 2-mm set-up margin. At treatment, 4.6 ± 1.5 breath-hold CBCTs were acquired for each patient and each fraction. The first CBCT of each fraction was used as the reference to quantify the geometric and dosimetric inter-breath hold variations on the subsequent CBCT sets in the treatment planning system. The geometric variation of the GTV was quantified as the variation of fiducial centroids. The metrics used for the dosimetric variation were GTV D95 ≥ Rx, PTV D95 ≥ Rx, PTV Dmax < 4290 cGy (130% Rx), Paddick conformity index (pCI), and the ratio of the volume of the 50% isodose volume and the PTV volume (R50%) for tumor, V20Gy < 20% for duodenum, stomach and bowel, V12% < 50% for liver, and V12% < 25% for combined kidneys, in addition to Dmax for each OAR. The tumor position showed an average variation of 0.0 ± 1.6 mm, -0.3 ± 1.5 mm, and -0.4 ± 2.6 mm in the LR, AP, and SI directions, respectively. The GTV D95, PTV D95, and PTV Dmax were changed by 0.8 ± 5.3%, -3.1 ± 14.0%, and -0.1 ± 3.5%. The pCI and R50% varied by 1.7 ± 6.5% and 1.3 ± 10.8%. The V20Gy of duodenum, stomach, and bowel varied by -0.2 ± 3.1%, -0.3 ± 2.5%, and -0.2 ± 4.8%. There was no sizable difference with V12% for liver and combined kidneys. The Dmax for duodenum, stomach, bowel, liver, and combined kidneys were changed by -0.7 ± 4.8%, -0.4 ± 8.2%, 0.8 ± 5.6%, 0.4 ± 4.3%, and -0.6 ± 3.5%, respectively. Inter-breath hold variation in pancreatic tumor position affects tumor coverage, dose conformity, dose fall-off and OAR doses, more significantly for Dmax of duodenum, stomach and bowel which are closely located to the tumor. Future work is to evaluate whether a robust optimization can account for the inter-breath hold variation to minimize the corresponding dosimetric deviations.