ROS1-targeted tyrosine kinase inhibitors (TKIs) are the standard treatment for ROS1-rearranged cancers. We here investigated the therapeutic significance of multiple fusion partners and variable ROS1 genomic breakpoints.We retrieved 81 ROS1-rearranged cases from a clinical DNA-based next generation sequencing cohort and performed comprehensive analyses, including reverse transcription-polymerase chain reaction, RNA sequencing and immunohistochemistry staining. The obtained data were correlated with the clinical responses to ROS1-TKIs.ROS1 fusions with canonical and ∼20 rare fusion partners were identified. ROS1 breaks occurred at rare 12 exon/intron regions together with major breakpoints. High ROS1 expression correlated significantly with major breakpoints and canonical partners and low expression associated with rare breakpoints and non-canonical partners (P < 0.001). Cases with an intron 32 ROS1 breakpoint involved exclusion of exon 33 to generate an in-frame fusion, and canonical fusion partners showed a strong preference for an intron 33 to intron 32 breakpoint (P < 0.001). Significantly better progression-free survival rates (PFSR) were observed among first-line TKI-treated NSCLC patients with ROS1 breakpoints in introns 33 and 34, compared with intron 32 and rare locations (80 %, 33.3 %, and 0 %, respectively; P < 0.001). Patients with at least one major breakpoint or canonical partner also showed significantly better PFSR compared to those with both rare partners and non-major breakpoints (50.4 % vs 0 %, P < 0.001).Canonical fusion partners with introns 33 and 34 of ROS1 may be the most optimal predictors for ROS1-TKI benefit. Precise characterization of the variants in terms of ROS1 breakpoints could be important for patient stratification in ROS1-rearranged cancers.