Abstract Background Osteosarcoma is a very aggressive tumor with a propensity to metastasize and invade surrounding tissue. Identification of the molecular determinants of invasion and metastatic potential may guide the development of a rational strategy for devising specific therapies that target the pathways leading to osteosarcoma. Methods In this study, we used pathway-focused low density expression cDNA arrays to screen for candidate genes related to tumor progression. Expression patterns of the selected genes were validated by real time PCR on osteosarcoma patient tumor samples and correlated with clinical and pathological data. Results THBS3, SPARC and SPP1 were identified as genes differentially expressed in osteosarcoma. In particular, THBS3 was expressed at significantly high levels (p = 0.0001) in biopsies from patients with metastasis at diagnosis, which is a predictor of worse overall survival, event-free survival and relapse free survival at diagnosis. After chemotherapy, patients with tumors over-expressing THBS3 have worse relapse free survival. High SPARC expression was found in 51/55 (96.3%) osteosarcoma samples derived from 43 patients, and correlated with the worst event-free survival (p = 0.03) and relapse free survival (p = 0.07). Overexpression of SPP1 was found in 47 of 53 (89%) osteosarcomas correlating with better overall survival, event-free survival and relapse free survival at diagnosis. Conclusion In this study three genes were identified with pattern of differential gene expression associated with a phenotypic role in metastasis and invasion. Interestingly all encode for proteins involved in extracellular remodeling suggesting potential roles in osteosarcoma progression. This is the first report on the THBS3 gene working as a stimulator of tumor progression. Higher levels of THBS3 maintain the capacity of angiogenesis. High levels of SPARC are not required for tumor progression but are necessary for tumor growth and maintenance. SPP1 is not necessary for tumor progression in osteosarcoma and may be associated with inflammatory response and bone remodeling, functioning as a good biomarker.