Background: Gastric cancer is the 4th leading cause of cancer death worldwide with high incidence in East Asia. Despite recent developments in treatment options, gastric cancer survival rate remains low, partially attribute to tumor heterogeneity. Here we performed a comprehensive single cell analysis of malignant cells from gastric tumor samples. In addition to characterizing heterogenous transcriptional patterns, we developed a framework to identify HER2+ malignant cells, which may complement standard diagnostics for treatment decisions. Methods: Published and in-house generated 10x Genomics single cell RNAseq data in tumor and normal samples from gastric cancer patients were aggregated, including 64 tumor samples and 27 pre-cancerous samples and 12 adjacent normal samples. The R package Seurat (version 4.0) was applied for data QC and downstream analysis. High quality cells were clustered using highly variable genes and cell types were annotated by curated marker genes. To identify malignant epithelial cells, a gastric cancer signature was derived using matched tumor and normal samples from TCGA gastric cancer patients. Gastric tumor samples were also leveraged to develop a HER2 signature and TCGA breast tumor samples were used to validate the signature. HER2+ malignant cells were identified by re-clustering malignant epithelial cells using the HER2 signature. Results: 345,361 high-quality single cells were profiled, representing major cell types within tumor microenvironment. Among 59,861 epithelial cells, 22,770 malignant cells were identified with high expression of a gastric cancer signature. Copy number alterations (aneuploid) were highly enriched in inferred malignant cells compared with non-malignant cells (48.6% vs 1.4%, p<2.2×10^-16, Fisher’s exact test). Further clustering analysis identified 10 subclusters with divergent expression of cancer oncogenic pathways, highlighting the vast heterogeneity of tumor origin. A Her2 amplification signature was derived using TCGA gastric cancer cohort (AUC=0.99) and validated in breast cancer cohort (AUC=0.93). Applying this signature to malignant cells, HER2+ cells were identified in a majority of tumors, with various proportions. In an independent data set, the percentage of HER2+ malignant cells is highly concordant with clinical HER2+ status, indicating the potential use of single cell data to supplement HER2+ clinical testing for treatment selection. Conclusions: This meta-analysis of single cell RNAseq data provides a comprehensive snapshot of tumor heterogeneity in gastric cancers, with aberrance in multiple oncogenic pathways observed in patients. The HER2+ cell classifier developed here holds the potential to complement standard clinical testing and enable personalized medicine, warranting further validation.