Recently, a simple model for evaluating turbulent scalar flux in premixed flames was developed and validated using six experimental data sets obtained from flames stabilized in impinging jets (Sabelnikov and Lipatnikov, Combust. Sci. Technol. 183 , 588–613, 2011 ; Sabelnikov and Lipatnikov, Flow Turbulence Combust. 90 , 387–400, 2013 ). The model addresses the flamelet regime of premixed turbulent combustion and yields an algebraic expression for the mean velocity conditioned to unburned mixture, while turbulent scalar flux is evaluated substituting this conditioned velocity into the well-known Bray-Moss-Libby (BML) expressions. The present work aims at further assessment of the aforementioned model against two well-known 3D DNS databases obtained from statistically planar, 1D premixed turbulent flames characterized by various density ratios (7.53, 5.0, 3.3, and 2.5). For the highest density ratio, an excellent agreement between the model and DNS data was obtained. This result is particularly encouraging, because the experimental data used earlier to test the model are associated with approximately the same (7-8) density ratios. However, the DNS data obtained for lower density ratios indicate a trend, not addressed by the original model, i.e. a model parameter is not a constant but decreases with decreasing density ratio, with the dependence of the model parameter on the density ratio being roughly linear for three flames addressed by one DNS database. Implementation of this linear fit into the model makes it consistent both with the DNS and with all experimental data used earlier to validate the original model.