Purpose: Radiomics has shown great promise to use quantifiable imaging characteristics to predict the behavior and prognosis of neoplasms. This is the first study to evaluate whether radiomic texture analysis can predict outcomes in gastroesophageal junction adenocarcinoma (GEJAC) treated with neoadjuvant chemoradiotherapy (CRT). Materials and Methods: Pretreatment contrast-enhanced CT images of 146 patients with stage II-III GEJAC were reviewed (2009-2011), and randomly split into training and validation groups at a 1:1 ratio stratified with baseline clinical characteristics. Whole-tumor texture was assessed using quantitative image features based on intensity, shape, and gray-level co-occurrence matrix. The relevant pretreatment texture features, in addition to the significant baseline clinical features to predict survival, were identified using multivariate Cox proportional hazard regression model with stepwise variable selection in the training sample and verified in the validation sample, to facilitate the proposal of a multi-point index for standard patient pre-treatment risk classification. Results: Of the factors identified in the training cohort independently associated with OS, only shape compactness (p = 0.04) and pathologic grade differentiation (PDG) (p = 0.02) were confirmed in the validation sample. Using both parameters, we created a 3-point risk classification index: low-risk (wellmoderate PDG and high compactness), medium-risk (poor PDG or low compactness), and high-risk (poor PDG and low compactness). The risk index showed a strong negative association with postoperative pathologic complete response (pCR) (p = 0.04). Median OS for the high-, medium-, and low-risk groups were 23, 51, and ≥ 72 months, respectively (p < 0.01). Similar results were seen with progression-free survival (respective 5-year rates of 15%, 30%, and 63%). Conclusion: Radiomic texture analysis can be used to stratify patients with GEJAC receiving trimodality therapy based on prognosis. The risk scoring system based on shape compactness and PDG shows a great potential for pre-treatment risk classification to guide surgical resection in locally advanced disease. Though in need of greater validation, these hypothesis-generating data could provide a unique platform of personalized oncologic care. © 2017 The Authors.