The diagnosis of follicular thyroid carcinoma (FTC) in the absence of metastasis can only be established postoperatively. Moreover, high-risk FTCs are often not identifiable at the time of diagnosis. In this study, we aimed to identify transcriptional markers of malignancy and high-risk disease in follicular thyroid tumors. The expression levels of 26 potential markers of malignancy were determined in a panel of 75 follicular thyroid tumors by a TaqMan quantitative RT-PCR approach. Logistic regression analysis (LRA) was used for gene selection and generation of diagnostic and prognostic algorithms. An algorithm based on the expression levels of five genes (TERT, TFF3, PPARgamma, CITED1, and EGR2) could effectively predict high-risk disease with a specificity of 98.5%. The metastatic potential could be predicted in all four cases with apparently benign or minimally invasive (MI) disease at the time of diagnosis, but poor long-term outcome. In addition, a second model was produced by implementing two genes (TERT and TFF3), which was able to distinguish adenomas from de facto carcinomas. When this model was tested in an independent series of atypical adenomas (AFTA) and MI-FTCs, 16 out of 17 AFTAs were classified as 'benign', while MI-FTCs with vascular invasion (sometimes referred to as 'moderately invasive') and/or large tumor size tended to classify in the 'malignant' group. The reported models can be the foundation for the development of reliable preoperative diagnostic and prognostic tests that can guide the therapeutic approach of follicular thyroid neoplasms with indeterminate cytology.