The relationship between glass structure and crystallization behavior has remained elusive, though a general solution would have broad applications. In this work, three-component glasses were studied in the system MTSixO2x+2, where M = Na, Li, or Ca, T = Al, B, or Fe, and x = 1, 2, and 3. These glasses, selected to systematically explore network and modifier substitutions, included mineralogically important silicate chemistries (i.e., albite, jadeite, carnegieite, nepheline, anorthite, eucryptite, spodumene, aegirine, reedmergnerite, malinkoite). Glasses were interrogated using neutrons and synchrotron X-rays, and empirical potential structure refinement (EPSR) modeling was performed to compare the average glass structures to the analogous crystalline references. Through EPSR modeling, it was shown that similarity in the coordination number of the non-network cations in glass systems (i.e., alkali and alkaline earth) predicts which glasses easily crystallize an isochemical phase during cooling from melt. Average coordination number of the M and T type cations of glass versus isochemical crystal is proposed as a metric to predict crystallization. This metric is compared to other proposed metrics such as normalized temperatures, change in density, fragility, and thermodynamic parameters. This the first study to employ EPSR modeling to relate crystallization propensity of isochemical crystals to network modifier short-range order.