Background Inclusion of trastuzumab in chemotherapy regimens is advantageous for patients with advanced or metastatic gastric cancer who overexpress HER2. Therefore, accurate assessment of HER2 status in tumor tissue is critical when weighing treatment options. Methods We examined HER2 expression in 180 paired endoscopic biopsy and surgical excision specimens of gastric cancers via immunohistochemistry (IHC). Equivocal IHC results (IHC 2?) were resolved by HER2 fluorescence in situ hybridization (FISH). The relationships of several clinical and pathological features with discordant HER2 results in paired specimens were determined. Results Fourteen biopsy specimens and surgical specimens (7.8 %) were HER2-positive. Discordant HER2 IHC scores were observed in 90 paired specimens (50 %) and 8 paired specimens (4.4 %) had discordant results. The kappa coefficients for an HER2 diagnostic algorithm were 0.264, 0.339, and 0.690 for IHC scores, IHC categories, and final results, respectively (p \ 0.001). Discordant HER2 results were significantly associated with discordant tumor differentiation in the paired biopsy and excision specimens (p = 0.01). Intratumoral heterogeneity did not predict HER2 discordance. There was no association between HER2 discordance and the number of biopsy tissue fragments (p = 0.764). Conclusions Hofmann's HER2 scoring system is a fairly reliable tool for evaluating HER2 status in biopsy and excision specimens. Discordant HER2 results in paired specimens were observed in a small percentage of gastric cancers. Testing all available specimens should be considered in order to eliminate discrepancies, especially when discordant tumor differentiation is observed.