A cohort including all female workers born 1906 through 1945 (n = 413,877) in Finland was identified through the Population Census of Finland of 1970. Incident cases of cancers of the gastrointestinal tract were explored during 1971 to 1995. Job titles in census records were converted to exposures of 31 occupational agents through a job-exposure matrix. For each agent, the product of level and probability of exposures was calculated and subdivided in three categories: zero, low and medium/high. Poisson regression models estimated relative risks (RR) for each agent, standardized for birth cohort, follow-up period, and socioeconomic status. Adjustment at job title level was done for alcohol use for cancers of the esophagus and liver and smoking for pancreatic cancer. The results showing either statistically significant RR at the medium/high level of exposure (RRH) or statistically significant trend (P < 0.05) over the exposure categories were considered as positive findings. Colon cancer risk (2009 cases) was positively associated with sedentary work (RRH 1.3, 95% CI = 1.1-1.6; P trend 0.001) and negatively associated with perceived workload (P trend = 0.007). For stomach cancer (1881 cases), we observed an association with exposure to electromagnetic fields (RRH 1.44, 95% CI = 1.01-2.05) and man-made vitreous fibers (MMVF) (p trend 0.03). Rectal cancer (1323 cases) showed an association with chromium (RRH 1.9, 95% CI = 1.2-3.1) and oil mist (RR 2.0; 95% CI = 1.0-3.9). For pancreas cancer (1302 cases) we found associations with exposure to chromium (RRH 1.8; 95% CI = 1.0-3.1; P trend 0.01), electromagnetic fields (RRH 1.8; 95% CI = 1.2-2.8; P trend 0.02), and sedentary work (RRH 1.3; 95% CI = 1.0-1.7; P trend 0.05). We found no significant associations between any FINJEM agents and cancers of the esophagus (389 cases), liver (389 cases), and gallbladder (651 cases). Having examined the associations between seven cancer sites and over 30 exposures there exists the real possibility that some of the associations detected are chance findings. Therefore, the associations observed should need to be confirmed in other studies.