The purpose of the present study was to estimate and compare the effects of macronutrient composition (relative portions of macronutrients) and of non-macronutrient components (e.g. water and fibre) on energy density (energy per unit weight) of the diets of human subjects. We used standard macronutrient energy content values to develop a simple conceptual model and equation for energy density in terms of % energy from dietary fat and % non-macronutrients by weight. To study these effects in self-selected diets of free-living subjects, we used four consecutive days of self-weighed and recorded food records for thirty-two male and thirteen female free-living adult subjects. In the range of typical human diets, the effect of % non-macronutrients by weight was several times greater than that of % energy from dietary fat, both in absolute terms and relative to daily variation in subjects' diets. Both effects were large enough to be physiologically important. Non-macronutrients (% by weight) alone explained much more of the variation in self-selected dietary energy density either between subjects (R2 95 %) or day-to-day (R2 95 %) than did % energy from dietary fat (R2 5 % and 6 % respectively). Omitting beverages gave similar results. The smaller effect of macronutrient composition on energy density of diets is mainly because alterations in macronutrient composition affect only the portion of typical dietary intake that is macronutrients (one-quarter to one-third of weight). Mathematical methods are also useful in analysing observational data and for separating effects of macronutrient composition and non-macronutrients in intervention studies. These results illustrate the importance of considering non-macronutrients in the design and analysis of experimental or observational dietary data.