LES models for turbulent non-premixed combustion usually require knowledge of the filtered density function of the conserved scalar, and we propose to use a simple top-hat function. Such top-hat distributions were developed as probability density functions for RANS applications in the 1970s but were soon surpassed by the β function. We find that in the context of LES, the top-hat distribution provides an excellent alternative to the now much more common β function. The top-hat function is assessed through a phenomenological analysis of Direct Numerical Simulation (DNS) data from a planar jet and of experimental data from a turbulent opposed jet. The approach is then tested a posteriori for a piloted diffusion flame (Sandia Flame D). Advantages of the top-hat function are the ease of implementation and the reduced dimensionality of look-up tables. The present paper also discusses inconsistencies of sub-grid β-FDFs, the FDFs sensitivity on implicit filtering, and the regime in which a β assumption can be a valid filtered density function for LES.