To identify a predictive molecular "signature" for sensitivity to retinoic acid in pancreatic cancer. Fourteen patient-derived, low-passage pancreatic ductal adenocarcinoma (PDAC) lines with varied expression of fatty acid-binding protein 5 (FABP5) and cellular retinoic acid-binding protein 2 (CRABP2) were used to evaluate the response to all-trans retinoic acid (ATRA). Cell proliferation, apoptosis, and migration/invasion assays were used to measure the in vitro response. Tumor growth was monitored in subcutaneous xenografts in athymic nude mice for 4 weeks. Response to ATRA was observed to be dependent upon differential expression of FABP5 versus CRABP2. Thus, elevated FABP5 expression was associated with minimal cytotoxicity and tumor growth inhibition and a paradoxical increase in migration and invasion. Conversely, CRABP2 expression in the absence of FABP5 was associated with significant tumor growth inhibition with ATRA, even in gemcitabine-resistant tumors. The ATRA-resistant phenotype of FABP5(high)CRABP2(null) cells could be circumvented by ectopic expression of CRABP2. Alternatively, reexpression of endogenous CRABP2 could be enabled in FABP5(high)CRABP2(null) PDAC lines by exposure to decitabine and trichostatin A, thereby relieving epigenetic silencing of the CRABP2 gene promoter. Immunohistochemical staining for FABP5 in archival human tissue microarrays identifies a subset of cases (13 of 63, ~20%) which are negative for FABP5 expression and might be candidates for ATRA therapy. The widely used agent ATRA deserves a "second look" in PDAC, but needs to be targeted to patient subsets with biopsy-proven FABP5-negative tumors, or be combined with a chromatin-modifying agent to reexpress endogenous CRABP2. © 2011 AACR.