Various computer simulations of molten glass flow have been used for the design, operation and trouble shooting of glass tank furnaces. It is necessary to predict the degree of thermal convection of molten glass, whose viscosity is affected by the major composition of glass and thermal conductivity of which varies with the ferrous oxide (FeO) content. The thermal conductivity decreases as the FeO content increases. Thermal convection is characterized by the Rayleigh number, Ra, and the Prandtl number, Pr. Typically, Ra is expected to be large for a small thermal conductivity, but the actual thermal convection is weaker in this case than when the thermal conductivity is large. To elucidate the underlying convection mechanism, a two-dimensional mathematical flow model is developed in this study. Non dimensional formulation of the governing equations shows that Pr-1 affects the degree of thermal convection. The effective value of Ra, Ra-eff, is introduced to evaluate the influence of temperature-dependent properties on thermal convection. An evaluation method of thermal convection is proposed and extended to glasses with different thermal conductivities.