AIM: To study the effects of phosphorus-32 glass microspheres (32P-GMS) on human hepatocellular carcinoma in nude mice.METHODS: Human liver cancer cell line was implanted into the dorsal subcutaneous tissue of 40 BALB/c nude mice.Then the 40 tumor-bearing BALB/c nude mice were allocated into treatment group (n=-32) and control group (n=8). In the former group different doses of 32P-GMS were injected into the tumor mass, while in the latter nonradioactive 31PGMS was injected into the tumor mass. The experimental animals were sacrificed on the 14th day. The ultrastructural changes of tumor in both treatment group and control group were studied with transmission electron microscopy (TEM)and stereology.RESULTS: In treatment group, a lot of tumor cells were killed and the death rate of tumor cells was much higher (35-70%). Ultrastructurally, severe nuclear damage was observed in the death cells. The characteristics of appoptosis such as margination of heterochromatin was also found in some tumor ceils. Besides, well differentiated tumor cells,degenerative tumor cells and some lymphocytes were seen.The skin and muscle adjacent to the tumor were normal. In control group, the tumor consisted of poorly differentiated tumor cells, in which there were only a few of dead cells (5%). Stereologicl analysis of ultrastructral morphology showed that Vv of nuclei (53.31±3.46) and Vv of nucleoli (20.40±1.84) in the control group were larger than those (30.21±3.52 and 10.96±2.52) in the treatment group respectively (P＜0.01), and Vv of RER (3.21±0.54) and Vv of mitochondria (4.53±0.89) in the control group were smaller than those (8.67±1.25 and 7.12±0.95) in the treatment group respectively (P＜0.01, 0.05). Sv of the membrane of microvilli and canaliculi (27.12 um2/100 um3±11.84 um2/100 um3) in the control group was smaller than that (78.81 um2/100 um3±19.69 um2/100 um3) in the treatment group (P＜0.01). But Vv of lipid particles (3.71±1.97) and Vv of vacuoles (5.72±1.58)were much larger than those (0.30±0.16 and 0.35±0.15) in the treatment group respectively (P＜0.05, P＜0.01).CONCLUSION: The experimental results indicate that local administration of 32P-GMS can produce obvious effect on liver cancer cells and the anticancer effect of 32P-GMS is directly proportional to the dose administrated. Ultrastructural stereology can also show the effect of 32P-GMS on the normalization of tumor cells, which is beneficial to the prognosis and treatment of patients. Moreover, local administrtion of 32P-GMS is also safe.