Genetic differences inferred from sequencing reads can be used for demultiplexing of pooled single-cell RNA-seq (scRNA-seq) data across multiple donors without WGS-based reference genotypes. However, such methods could not be directly applied to single-cell ATAC-seq (scATAC-seq) data owing to the lower read coverage for each variant compared to scRNA-seq. We propose a new software, scATAC-seq Variant-based EstimatioN for GEnotype ReSolving (scAVENGERS), which resolves this issue by calling more individual-specific germline variants and using an optimized mixture model for the scATAC-seq. The benchmark conducted with three synthetic multiplexed scATAC-seq datasets of peripheral blood mononuclear cells and prefrontal cortex tissues showed outstanding performance compared to existing methods in terms of accuracy, doublet detection, and a portion of donor-assigned cells. Furthermore, analyzing the effect of the improved sections provided insight into handling pooled single-cell data in the future.