Background: Cancer of unknown primary (CUP) constitutes 3%-5% of all newly diagnosed cancer cases. It presents a major diagnostic challenge as knowing the tumor tissue of origin (ToO) of the cancer is crucial for choosing the optimal treatment. MicroRNAs are a family of non-coding, regulatory RNA genes involved in development and differentiation that were shown to be involved in carcinogenesis. MicroRNAs, highly stable in clinical samples and tissue-specific, serve as ideal biomarkers for cancer diagnosis. Our first generation assay identifies the ToO using a set of 48 microRNAs measured on a qRT-PCR platform and differentiates 25 different tumor types. We present here the development and validation of a second generation clinical assay that can identify 42 different tumor types using a custom array platform. Methods: Over 1300 primary and metastatic tumor formalin-fixed and paraffin-embedded (FFPE) tissue samples were used for the training and development of the assay. High-purity RNA was extracted using proprietary protocols. Expression levels of all known and additional Rosetta’s proprietary microRNAs were profiled using a custom array platform. A combination of a decision tree with a KNN classifier was developed to identify the ToO based on the expression of 64 microRNAs. A validation set of more than 500 independent FFPE tissue samples of primary tumors and metastases was analyzed blindly and classified using the developed assay. Results: The assay is able to identify 42 tumor types that include carcinomas, soft tissue tumors, lymphoma and other malignancies with very high accuracy. Based on the validation set, the sensitivity (positive agreement) for identifying the tumor origin is 85%. In more than 80% of the cases, a single answer is reported, and for these cases, the sensitivity is 90%. Conclusions: Previous studies have highlighted the tissue-specificity of microRNA expression, and have demonstrated their potential use for classification of human malignancies. An enlarged panel of 42 ToO that can be identified with high accuracy promises improved utility for the diagnosis of cancers of unknown/uncertain primary.