A nonlinear low-Reynolds-number heat transfer model is developed to predict turbulent flow and heat transfer in separated and reattaching flows. The $k-{\varepsilon}-f_{\mu}$ model of Park and Sung (1997) is extended to a nonlinear formulation, based on the nonlinear model of Gatski and Speziale (1993). The limiting near-wall behavior is resolved by solving the $f_{\mu}$ elliptic relaxation equation. An improved explicit algebraic heat transfer model is proposed, which is achieved by applying a matrix inversion. The scalar heat fluxes are not aligned with the mean temperature gradients in separated and reattaching flows; a full diffusivity tensor model is required. The near-wall asymptotic behavior is incorporated into the $f_{\lambda}$ function in conjunction with the $f_{\mu}$ elliptic relaxation equation. Predictions of the present model are cross-checked with existing measurements and DNS data. The model preformance is shown to be satisfactory.