Plasma transforming growth factor beta1 (TGF-beta1) has been reported to be correlated with the extent of disease in colorectal cancer, but it is not known whether measuring this cytokine can help predict liver metastasis after curative resection. We prospectively studied whether plasma TGF-beta1 levels could predict liver metastasis in 117 patients with colorectal cancer before and after curative resection. Blood samples were drawn before and 2 weeks after surgery to determine the cytokine levels. Abdominal ultrasonography or computed tomography was done every 3 months after surgery. The primary end point for follow-up was recurrence. Seventy-seven of 117 cases (66%) had preoperative levels of the cytokine higher than the borderline limit of 7.5 ng/ml. Postoperative levels were >7.5 ng/ml in 29 of 117 patients (25%). The median follow-up period was 42 months (range, 5--66 months), with follow-up of all 117 patients. No recurrence was observed in 13 patients with Dukes' stage A lesions. Liver metastasis occurred in 18 of 104 patients (17%) with Dukes' stage B or C disease. Fourteen of 18 patients (78%) who developed liver metastasis had shown a postoperative plasma TGF-beta1 level of >7.5 ng/ml. Cox proportional hazards regression analysis showed that the postoperative level was a significant predictive factor for liver metastasis (P < 0.001). A single point measurement of plasma TGF-beta1 levels at 2 weeks after curative resection seems to be able to predict liver metastasis in colorectal cancer. This finding suggests the value of a prospective trial of liver-targeted adjuvant therapy for patients with elevated postoperative plasma TGF-beta1 levels.