Abstract Functionality of training systems different in size, shape and geometry are primarily a function of their ability to intercept and distribute light effectively within the canopy. In peach, methodologies for a rapid and reliable assessment of such features are still lacking. In this study we propose a systemic approach that as unique data entry requires diurnal ground monitoring of the light-shadow windows of a tree canopy. Case studies for canopy shapes were a pyramid (▴, Delayed Vase), a parallelogram (♦, Palmette) and a Y (Tatura trellis) chosen within a 3-year-old peach orchard. Canopy geometrical and structural parameters calculated from above and below canopy radiation readings taken at full canopy development include Silhouette ( S ) as sunlit canopy area projected orthogonal to the sunbeam, leaf layer index (LLI), canopy leaf projection coefficient computed orthogonal to sunbeam direction ( K ⌋ ), instantaneous canopy photon influx ( Q CA ), instantaneous canopy intercepted photon flux in the 300–1100 nm waveband ( Q C ) and canopy photon influx capture efficiency ( e Q CA → Q C ). Whole-tree gas exchange was also continuously monitored for a week on each canopy shape to gain a direct measurement of canopy net assimilation rate ( A C ) and canopy transpiration rate ( E C ). A positive Q C vs. A C correlation was shown by any canopy type, with r = 0.93, 0.97 and 0.92 for ▴, ♦ and Y , respectively. By contrast, while Q C and E C were weakly correlated in ▴ and ♦, a close positive correlation ( r = 0.87) was found between these two variables in Y . The Tatura trees also showed, regardless of timing of the day, the highest E C / A C , hence better water use efficiency. This study validates the hypothesis that a systemic assessment of canopy quantum flux absorption ( Q C ) leads to reliable prediction of actual net canopy photosynthetic rates paving the way to: (a) easier and faster evaluation of efficiency of canopy systems differing in size and shape and (b) simplification in whole-canopy photosynthetic models.