Background: Carcinoma of unknown primary (CUP) has a poor prognosis and accounts for diagnoses in > 30,000 patients annually in the United States. Current tissue of origin (TOO) methods use protein or RNA-based expression signatures, are costly, and often provide little clinically relevant information. We previously showed that use of comprehensive genomic profiling (CGP) identifies a clinically relevant genomic alteration in 96% of CUPs. We hypothesized that CGP could also potentially identify a TOO for some CUP patients, further refining potential targeted therapeutic options. Methods: DNA alterations from 44,065 samples that underwent CGP as part of standard clinical care were classified using logistic regression based on mutational profiles and select clinical features. Results: We first trained a logistic regression classifier to identify non-small cell lung cancers (NSCLC) with alterations in EGFR, ERBB2, BRAF, MET, KRAS, ALK, RET, or ROS1 (lung gene set). In cross-validation, the classifier correctly identified 79% of NSCLCs (92% of EGFR mutant samples) and 93% of non-NSCLCs. When applied to a set of 1352 CUPs with an alteration in the lung gene set, 425 (31%) were predicted to be NSCLC. We next trained a classifier to distinguish BRAF V600-mutant colorectal cancers (CRCs) from other BRAF V600-mutant cancers. The classifier correctly identified 90% of CRCs and 93% of non-CRC. Use of this classifier with a BRAF-mutant CUP data set predicted 32/87 (37%) of BRAF-mutant CUP tumors to be CRC. Conclusions: In a large series of CUP patients with alterations in the lung gene set, one-third were likely NSCLC, and could potentially benefit from approved targeted therapies in this disease. When applied to BRAF V600-mutant CUPs, 37% were predicted to be CRC, and therefore single agent vemurafenib therapy is unlikely to be effective and combination therapy indicated. This approach is useful for identifying targeted therapeutics likely to be effective based on molecular match and TOO. We plan to extend this DNA-based tool to 4 additional tumor types. Application of this classifier to CUPs may help identify therapeutic strategies from CGP without the need for additional material or testing.