Anterior gradient 2 (AGR2) is a gene predominantly expressed in mucus-secreting tissues or in endocrine cells. Its expression is drastically increased in tumors including prostate cancer. Here we investigated whether AGR2 transcript levels can be used as a biomarker to detect prostate cancer (PCa). Using a PCR-based approach, we could show that in addition to the wild-type (AGRwt long and short) transcripts, five other AGR2 splice variants (SV) (referred to as AGR2 SV-C, -E, -F, -G and -H) were present in cancer cell lines. In tissue biopsies, SV-H and AGR2wt (short) distinguished between benign and PCa (p ≤ 0.05 n = 32). In urine exosomes, AGR2 SV-G and SV-H outperformed serum PSA. Receiver operating characteristic (ROC) curves showed the highest discriminatory power of SV-G and SV-H in predicting PCa. AGR2 SV-G and SV-H are potential diagnostic biomarkers for the non-invasive detection of PCa using urine exosomes.