Importance: Although profiling of gene expression and gene alterations by next-generation sequencing (NGS) to predict the primary tumor site and guide molecularly targeted therapy might be expected to improve clinical outcomes for cancer of unknown primary site (CUP), to our knowledge, no clinical trial has previously evaluated this approach. Objective: To assess the clinical use of site-specific treatment, including molecularly targeted therapy based on NGS results, for patients with CUP. Design, Setting, and Participants: This phase 2 clinical trial was conducted at 19 institutions in Japan and enrolled 111 previously untreated patients with the unfavorable subset of CUP between March 2015 and January 2018, with 97 patients being included in the efficacy analysis. Eligibility criteria included a diagnosis of unfavorable CUP after mandatory examinations, including pathological evaluation by immunohistochemistry, chest-abdomen-pelvis computed tomography scans, and a positron emission tomography scan. Interventions: RNA and DNA sequencing for selected genes was performed simultaneously to evaluate gene expression and gene alterations, respectively. A newly established algorithm was applied to predict tumor origin based on these data. Patients received site-specific therapy, including molecularly targeted therapy, according to the predicted site and detected gene alterations. Main Outcomes And Measures: The primary end point was 1-year survival probability. Secondary end points included progression-free survival (PFS), overall survival (OS), objective response rate, safety, efficacy according to predicted site, and frequency of gene alterations. Results: Of 97 participants, 49 (50.5%) were women and the median (range) age was 64 (21-81) years. The cancer types most commonly predicted were lung (21 [21%]), liver (15 [15%]), kidney (15 [15%]), and colorectal (12 [12%]) cancer. The most frequent gene alterations were in TP53 (45 [46.4%]), KRAS (19 [19.6%]), and CDKN2A (18 [18.6%]). The 1-year survival probability, median OS, and median PFS were 53.1% (95% CI, 42.6%-62.5%), 13.7 months (95% CI, 9.3-19.7 months), and 5.2 months (95% CI, 3.3-7.1 months), respectively. Targetable EGFR mutations in tumor specimens were detected in 5 patients with predicted non-small-cell lung cancer (5.2%), 4 of whom were treated with afatinib; 2 of these patients achieved a durable PFS of longer than 6 months. Conclusions and Relevance: This study's findings suggest that site-specific treatment, including molecularly targeted therapy based on profiling gene expression and gene alterations by NGS, can contribute to treating patients with the unfavorable subset of CUP. Trial Registration: UMIN Identifier: UMIN000016794.