Mesenchymal stromal cells (hMSCs) are advancing into the clinic but the therapeutic efficacy of hMSCs faces the problem of donor variability. In bone tissue engineering, no reliable markers have been identified which are able to predict the bone-forming capacity of hMSCs prior to implantation. To this end, we isolated hMSCs from 62 donors and characterized systematically their in vitro lineage differentiation capacity, gene expression signature and in vivo capacity for ectopic bone formation. Our data confirms the large variability of in vitro differentiation capacity which did not correlate with in vivo ectopic bone formation. Using DNA microarray analysis of early passage hMSCs we identified a diagnostic bone-forming classifier. In fact, a single gene, CADM1, strongly correlated with the bone-forming capacity of hMSCs and could be used as a reliable in vitro diagnostic marker. Furthermore, data mining of genes expressed correlating with in vivo bone formation represented involvement in neurogenic processes and Wnt signaling. We will apply our data set to predict therapeutic efficacy of hMSCs and to gain novel insight in the process of bone regeneration. Our bio-informatics driven approach may be used in other fields of cell therapy to establish diagnostic markers for clinical efficacy.