We reported two Au clusters with precisely controlled molecular size (Au(5)Peptide(3) and Au(22)Peptide(10)) showing different antitumor effects. In vitro, both Au(5)Peptide(3) and Au(22)Peptide(10) were well taken up by human nasopharyngeal cancer cells (CNE1 cells). However, only Au(5)Peptide(3) significantly induced CNE1 cell apoptosis. Further studies showed that CNE1 cells took up Au(5)Peptide(3) (1.98 X 10(-15) mol/cell), and 9% of them entered mitochondria (0.186 X 10(-15) mol/cell). As a comparison, the uptake of Au(22)peptide(10) was only half the amount of Au(5)Peptide(3) (1.11 X 10(-15) mol/cell), and only 1% of them entered mitochondria (0.016 X 10(-15) mol/cell). That gave 11.6-fold more Au(5)Peptide(3) in mitochondria of CNE1 cells than Au(22)Peptide(10). Further cell studies revealed that the antitumor effect may be due to the enrichment of Au(5)Peptide(3) in mitochondria. Au(5)Peptide(3) slightly decreased the Mcl-1 (antiapoptotic protein of mitochondria) and significantly increased the Puma (pro-apoptotic protein of mitochondria) expression level in CNE1 cells, which resulted in mitochondrial transmembrane potential change and triggered the caspase 9 caspase 3 PARP pathway to induce CNE1 cell apoptosis. In vivo, CNE1 tumor growth was significantly suppressed by Au(5)Peptide(3) in the xenograft model after 3 weeks of intraperitoneal injection. The TUNEL and immuno-histochemical studies of tumor tissue verified that CNE1 cell apoptosis was mainly via the Puma and Mcl-1 apoptosis pathway in the xenograft model, which matched the aforementioned CNE1 cell studies in vitro. The discovery of Au-5 but not Au-22 suppressing tumor growth via the mitochondria target was a breakthrough in the nanomedical field, as this provided a robust approach to turn on/off the nanoparticles' medical properties via atomically controlling their sizes.