Neuroblastoma exhibits two cellular phenotypes: therapy-sensitive adrenergic (ADRN) and therapy-resistant mesenchymal (MES). To understand treatment response, it is important to elucidate how these phenotypes impact the dynamics of cancer cell populations and whether they represent distinct cell types or dynamic cell states. Here, we use an integrated experimental and mathematical modeling approach. We experimentally measure the fractions of ADRN and MES phenotypes under baseline (untreated) conditions and under repeated treatment cycles. We develop evolutionary game theoretic models predicting how the populations would respond if ADRN and MES phenotypes (1) are distinct cell types or (2) represent dynamic cell states and fit these models to the experimental data. We find that, although cells may undergo an ADRN to MES phenotypic switch under treatment, the best-fit model sees ADRN and MES as distinct cell types. Differential proliferation and survival of these two cell types, and not cell-state switching, drive therapeutic response.