Engineered tree fruit architecture is a result of multiple techniques, which include training, pruning, and thinning. The tree architecture influences light interception, canopy growth, fruit quality, and yield, and in addition to allows efficient orchard management (e.g. trimming and harvesting). In this work, image-based techniques were evaluated to phenotype architectural traits of apple trees. A digital red-green-blue (RGB) camera was used in proximal sensing to collect apple tree side-images, while another RGB camera mounted on an unmanned aerial system (UAS) was utilized in remote sensing (15 m above ground level) to collect data from tree canopy. Ground reference data were collected from trial blocks of ‘WA38’ apple trees, which included three training systems (Spindle, V-trellis, and Bi-axis), two rootstocks (G41 and M9-Nic29) and two pruning methods (Bending and Click), and were compared to image-based traits. Architectural traits were extracted from 2D images (proximal) and 3D digital surface model (remote sensing), including box-counting fractal dimension (DBs), middle branch angle, number of branches, trunk basal diameter, and tree row volume. Significant (P < 0.0001) difference in DBs between Spindle and V-trellis training systems was observed, while significant (P < 0.05) correlations between DBs with tree height (r = 0.78) and total yield per unit area in Mton/hectare (r = 0.70) was found from proximal sensing results. Meanwhile, significant (P < 0.05) correlations between average or total tree row volume and ground reference data, such as trunk area, total fruit yield per tree, were detected from remote sensing results. Results from this study demonstrated the potential of image-based phenotyping techniques in evaluating tree architectural traits. We believe that such imaging techniques can assist tree fruit breeders, physiologists, and growers in evaluating tree architectural traits easily and efficiently.