Background: Sequential measurements of CA125 have been used in a longitudinal algorithm to assess the risk of ovarian cancer. Rising CA125 followed by transvaginal sonography can achieve high specificity (99.6%) and positive predictive value (36%), but as 20% of ovarian cancers do not express CA125, sensitivity will not be optimal for preclinical disease. Multiple biomarkers may have greater sensitivity than CA125 alone. We developed and evaluated an algorithm that estimates risk of ovarian cancer using multiple biomarkers measured annually. Methods: Serum levels of CA125, CA15.3, CA72.4, CA19.9, and HE4 were evaluated in samples from the MDACC SPORE normal risk ovarian screening study (NROSS) and the GOG. The NROSS included 5 annual samples from 197 women for whom ovarian cancer was not detected and 9 cases (3 invasive ovarian cancer, 3 LMP ovarian cancer, 3 benign). The GOG contained preoperative sera from 95 Stage I ovarian cancer cases. 100 randomly selected controls together with the surgically evaluated cases in NROSS were used for model derivation. The remaining 97 from NROSS and the GOG cases were used for validation. The biomarkers were combined into a composite index using the UMSA learning algorithm. A simple longitudinal algorithm was then applied to assess the risk of cancer using both the composite index and CA125 at 98% specificity so that no more than 2% of women screened would be referred to ultrasound. Results: At a weighted average annual specificity of 98% on the first 5 time points of the longitudinal samples, the model’s sensitivity in detecting the GOG stage I cases was 81.1% (77/95). This compares favorably to a sensitivity of 73.7% (70/95) using CA125 alone at the same 5 year weighted annual specificity (p < 0.02). In longitudinal use of the composite index, the algorithm was able to detect all the ovarian cancer cases (3 invasive, 3 LMP) among the 9 surgically evaluated cases while maintaining a yearly average specificity of 98%. Conclusions: Multiple biomarkers can be combined in a longitudinal algorithm to improve detection of early stage ovarian cancer. Additional sample sets that include independent samples of longitudinally detected cancer cases will be needed to further validate the algorithm.