Background: Cancer of unknown primary (CUP) with a “gastrointestinal (GI) profile” appears to be a clinically distinct subset, based on immunophenotyping, specifically staining with cytokeratins 20 (CK20) and 7 (CK7), a type I and II keratin, respectively and caudal type homeobox 2 (CDX2) protein, a transcription factor expressed in nuclei of intestinal epithelial cells. However, only a limited clinicomolecular account of this entity exists. Comprehensive profiling is needed to impact personalized therapeutics and improve prognosis. Methods: We identified 497 pts using a retrospective/prospectively managed CUP database and tumor registry of pts evaluated at MD Anderson Cancer Center from 2012-2016. Pts were classified into 3 cohorts based on immunohistochemistry: lower GI profile CUP (LGI-CUP) [CK20+/CK7- or CK20+/CK7+/CDX2+], upper GI profile CUP (UGI-CUP) [CK20-/CDX2+ or CK7+/CK20+/CDX2-] and remaining were treated as a non-GI CUP (NGI-CUP) control group [CK20-/CDX2-]. Clinical and pathological data including molecular profiling, therapy and survival were logged. Fisher-exact test was used. Kaplan-Meier method was used to estimate overall survival (OS) and compared with log-rank test. Results: Among 497 pts, 316 (63.6%) had adequate immunostaining for analysis. Of these, 76 (24.1%), 75 (23.7%) and 165 (52.2%) were classified as LGI-CUP, UGI-CUP, and NGI-CUP, respectively. Median age at diagnosis for the 3 cohorts was 59, 62 and 60 years, respectively. Key baseline clinicopathological characteristics were balanced between groups except histology and Culine risk stratification. LGI-CUP were enriched for adenocarcinoma (89% v 54% v 57%, P = 0.009) and good risk group (65% v 46% v 43%, P < 0.001) compared to UGI-CUP and NGI-CUP, respectively. Median OS for the 3 cohorts were: 18.5 months (95%CI: 10.8–26.2) for LGI-CUP compared to 12.3 months (95%CI: 6.3–18.3) for UGI-CUP (P = 0.06) and 13.5 months (95%CI: 10.6–16.6) for NGI-CUP (P = 0.040). On multivariate analyses, LGI-CUP subtype emerged as an independent prognostic factor for better overall survival (HR 0.69, 95%CI: 0.5–0.9, P = 0.046) in addition to histology and Culine risk group. Among LGI-CUP, 39 pts with complete treatment data, 29 (74%) received frontline colorectal cancer specific chemotherapy (5-FU/Capecitabine-based) therapy. Median time to treatment failure was 3.6 months compared 1.8 months for those who received non-colorectal therapy (P = 0.19). For LGI-CUP, most common genomic alterations were TP53 (22%), ERBB2 (12%), KRAS (10%). For UGI-CUP, most common genomic alterations were KRAS (11%), and TP53 (9%). Conclusions: CUP pts with lower GI profile have improved overall survival compared to those with an upper GI or a non-GI immunophenotype. Identifying these CUP subsets and rational use of immunotherapy and targeted therapy may provide benefit to CUP pts, particularly as these treatments evolve for the management of known GI cancers.