Anti-G(D2) murine antibody 3F8 plus subcutaneously (sc) administered granulocyte-macrophage colony-stimulating factor (GM-CSF) was used against primary refractory neuroblastoma in metastatic osteomedullary sites. Large study size and long follow-up allowed assessment of prognostic factors in a multivariate analysis not reported with other anti-G(D2) antibodies. In a phase II trial, 79 patients without prior progressive disease were treated for persistent osteomedullary neuroblastoma documented by histology and/or metaiodobenzyl-guanidine (MIBG) scan. In the absence of human antimouse antibody, 3F8 + scGM-CSF cycles were repeated up to 24 months. Minimal residual disease (MRD) in bone marrow was measured by quantitative reverse transcription-polymerase chain reaction pre-enrollment and post-cycle #2, before initiation of 13-cis-retinoic acid. Study endpoints were: (i) progression-free survival (PFS) compared with the predecessor trial of 3F8 plus intravenously administered (iv) GM-CSF (26 patients) and (ii) impact of MRD on PFS. Using all 105 patients from the two consecutive 3F8 + GM-CSF trials, prognostic factors were analyzed by multivariate Cox regression model. Complete response rates to 3F8 + scGM-CSF were 87% by histology and 38% by MIBG. Five-year PFS was 24 ± 6%, which was significantly superior to 11 ± 7% with 3F8 + ivGM-CSF (p = 0.002). In the multivariate analysis, significantly better PFS was associated with R/R or H/R FCGR2A polymorphism, sc route of GM-CSF and early MRD response. MYCN amplification was not prognostic. Complement consumption was similar with either route of GM-CSF. Toxicities were manageable, allowing outpatient treatment. 3F8 + scGM-CSF is highly active against chemoresistant osteomedullary neuroblastoma. MRD response may be an indicator of tumor sensitivity to anti-G(D2) immunotherapy. Correlative studies highlight the antineoplastic potency of myeloid effectors.