The term "gastrointestinal stromal tumor" (GIST) has been applied to a collection of distinctive mesenchymal tumors occurring within the human gastrointestinal tract. As new drug therapy becomes available, data regarding the natural history of these unusual tumors are necessary to provide selection factors for treatment. Ninety-eight patients had light microscopy compatible with GIST at a single institution from 1989 to 2000. After immunostaining with c-kit and histopathologic review, 69 were judged to be GIST. All prognostic indicators were determined for gastric GIST, intestinal GIST, and all locations combined. The location of the GIST did not have a significant impact on survival. Clinically, tumor size, peritoneal cancer index, and completeness of cytoreduction had a significant impact on prognosis for GIST at all locations. Pathologically, cytologic atypia, necrosis, invasion and number of mitoses were significant prognostic indicators for GIST. Criteria to separate three pathologic groups of GIST according to the tumor size and the mitotic count were useful to evaluate the tumor behavior; in the borderline pathologic group invasion and cytologic atypia were statistically significant prognostic criteria. The cell phenotypes, as determined by immunostains, correlated with the prognosis of gastric GIST but not intestinal GIST. A correlation between the immunostain Ki-67 but not CD-34 or desmin and the prognosis was observed. It is possible to select clinical and pathologic parameters of GIST that impact on prognosis. Invasion and necrosis help to determine the prognosis with borderline tumors. The immunostain Ki-67 correlated with the prognosis and may be helpful to assess prognosis when dealing with small biopsy specimens.