The heat of hydration of hydraulic cements depends on a complex set of phase dissolution and precipitation reactions following the addition of water. Heat of hydration is currently measured in one of two ways: acid dissolution of the raw cement and a hydrated cement after 7 days or isothermal calorimetry. In principle, the heat of hydration should be predictable from knowledge of the cement composition and perhaps some measure of the cement fineness or total surface area. The improved mineralogical estimates provided by quantitative X-ray powder diffraction, together with improved statistical data exploration techniques that examine nonlinear combinations of candidate model constituents, were used to explore alternative predictive models for the 7-day heat of hydration. An exploratory tool, called “all possible alternating conditional expectations,” was created by combining all possible subsets regression with alternating conditional expectation to judiciously select variables within an explanatory variable class and subsets of variables across explanatory variable classes exhibiting the highest potential predictive power for additive nonlinear models for 7-day heat of hydration. Although a single, strong model for 7-day heat of hydration did not emerge from analyses, general conclusions were drawn. Good-fitting models included a key structural mineralogical phase (belite preferred); calcium sulfate phase (bassanite preferred); total fineness or surface area component (Blaine fineness preferred); and ferrite in conjunction with iron oxide, or aluminate, or cubic aluminate.