Blood-based B-cell activating factor (BAFF), growth differentiation factor-15 (GDF-15) and osteopontin (OPN) have been identified to be promising biomarkers for the metastases of uveal melanoma (UM). This study intended to assess their kinetics and to evaluate their significance as a three-marker panel. A group of 36 UM patients with and 137 patients without metastases were included in the study. Their plasma OPN levels were measured by ELISA; serum BAFF and GDF-15 levels were determined with a Luminex MAGPIX system. Receiver operating characteristic (ROC) analysis was performed to calculate the cutoff values of the three markers for identifying the patients with metastases. The ability to identify patients with metastases was compared between the single markers and the combination as a three-marker panel. By using the Student’s t-test, we also investigated the kinetic changes of the levels of BAFF, GDF-15 and OPN across six periods (i.e., 0–6 months, 6–12 months, 12–18 months, 18–24 months, >24 months and post-metastasis) before the imaging diagnosis of metastases. By maximizing the Youden’s index, the serum GDF-15 level of 1209 pg/mL and the plasma OPN level of 92 ng/mL were identified to have the best performance for distinguishing the metastatic patients from non-metastatic patients. The three-marker panel offered a better performance in distinguishing patients with metastases, with an area under the curve of 0.802, than any single biomarker. Increasing trends of the levels of three biomarkers were observed in the two-year period before the imaging diagnosis of metastases. The combined panel of BAFF, GDF-15 and OPN might be a utilizable implementation for the detection of UM metastases. In the bioinformatics study with two external datasets, the high expression of gene BAFF and GDF-15 in primary UM tissues was identified to be associated with poor overall survival rates. As the current work is a single-center retrospective study, more well-designed prospective investigations employing larger cohorts are urgently needed to validate our findings.