Identification of patients with a poor prognosis for non-metastatic rhabdomyosarcoma (RMS) remains a clinical challenge. Prospective analysis for the presence of disseminated RMS cells in bone marrow at diagnosis, using immunocytochemistry, with MyoD1 and myogenin as markers, was carried out. Thirty-seven patients treated on RMS88 and RMS96 Italian protocols underwent staging investigations, and in addition marrow examination for occult tumour cells. All patients had negative marrow involvement using cytomorphology, but 10/37 were positive with immunostaining. With a median follow-up of 46 months (range, 12-115), 7 patients had died and 30 were disease-free. Overall survival probability was 92% in patients with no occult marrow infiltration, 47% with occult marrow infiltration (P=0.001); event-free survival probability was 89% in the former and 50% in the latter (P=0.01). Disseminated tumour cells are indicative of disease spread and are significantly linked to recurrence at distant sites and poorer outcome. Marrow examination at diagnosis using immunocytochemistry may be an additional tool to modulate treatment.