Abstract Lineage plasticity is a well–established mechanism of resistance to targeted therapies in lung and prostate cancer, where tumors transition from adenocarcinoma to small–cell or neuroendocrine carcinoma. Through single–cell analysis of a cohort of heavily–treated castration–resistant human prostate cancers (CRPC), we report a greater degree of plasticity than previously appreciated, with multiple distinct neuroendocrine (NEPC), mesenchymal (EMT–like), and other subpopulations detected within single biopsies. To explore the steps leading to this plasticity, we turned to two genetically engineered mouse models of prostate cancer that recapitulate progression from adenocarcinoma to neuroendocrine disease. Time course studies reveal expansion of stem–like luminal epithelial cells ( Sca1 +, Psca +, called L2) that, based on trajectories, gave rise to at least 4 distinct subpopulations, NEPC ( Ascl1 +), POU2F3 ( Pou2f3 +), TFF3 ( Tff3 +) and EMT–like ( Vim +, Ncam1 +)––these populations are also seen in human prostate and small cell lung cancers. Transformed L2–like cells express stem–like and gastrointestinal endoderm–like transcriptional programs, indicative of reemerging developmental plasticity programs, as well as elevated Jak/Stat and interferon pathway signaling. In sum, while the magnitude of multilineage heterogeneity, both within and across patients, raises considerable treatment challenges, the identification of highly plastic luminal cells as the likely source of this heterogeneity provides a target for more focused therapeutic intervention. One Sentence Summary Multilineage plasticity results from expansion of stem–like luminal cells with JAK/STAT activation, serving as a therapeutic target.