The unresolved scalar variance in large-eddy simulations of turbulent flows is a funda- mental physical and modelling parameter [1, 2]. Despite its importance, relatively few algebraic models have been developed for this important variable with the most prominent model to date being the classic gradient-based model. In this work, a general mathemat- ical framework is presented for constructing base static variance models. The proposed framework does not invoke any assumptions on the flow, the nature of the scalar, or the reaction regime. It is demonstrated that higher-order reconstructions of the primary resolved variables naturally lead to base static models of increased accuracy, and that the classic gradient model is a subset of a more general higher-order model. The clas- sic scale-similarity assumption for developing dynamic models is then re-visited, and it is demonstrated that this can be interpreted as a two-level reconstruction-modelling ap- proach. Based on this result, a general methodology is proposed for constructing dynamic models for any given general reconstruction operator and base static model. Consequently, improved static and dynamic models for the scalar variance are developed. The models are thoroughly tested a priori using two high-fidelity direct numerical simulation databases corresponding to two substantially different flame and flow configurations.