The increasing demand for water in developing countries, like India, requires efficient water management and resource allocation. This is crucial to accurately assess and predict hydrological processes such as streamflow, drought, and flood. However, simulations of these hydrologic processes from various hydrological models differ in their accuracy. By analyzing different characteristics of hydrological models, selection scores can be used to select the best model for the intended purpose based on their inherit strengths (i.e., some models are better for streamflow prediction). In this study, 13 different criteria were used for the model selection scores including temporal and spatial resolutions, and processes involved. Thereafter, based on different scores, we selected two different hydrological models for streamflow prediction in the Kangsabati River Basin (KRB) in eastern India, namely (1) Génie Rural à 4 paramètres Journalier (GR4J), a conceptual model, and (2) Variable Infiltration Capacity (VIC), a semi-distributed model. The models were calibrated against the daily observed streamflow at upper KRB (Reservoir) and lower KRB (Mohanpur) from 2000 to 2006 and validated during the period from 2008 to 2010. Despite the differences in model structure and data used, both models simulated streamflow at a daily time scale with Nash–Sutcliffe coefficient of 0.71–0.82 for the VIC model and 0.63–0.71 for the GR4J. Due to the simpler structure, parsimonious nature, fewer parameters, and reasonable accuracy, the results suggest that a conceptual rainfall—runoff model like GR4J can be used in data-deficient conditions.