Christopher W. Tyler
Smith-Kettlewell Eye Reseach Institute, San Francisco


Although the eyes are a key feature of facial portraits, compositional rules for the placement of the eyes relative to the frame seem to be lacking. Two hypotheses were evaluated for a relation between eye position and the portrait frame. One was that the pair of eyes might be placed symmetrically in portraits where both eyes are visible. This center of facial symmetry is often discussed in art analysis and should be an explicit primitive in portrait composition. The alternative hypothesis of a consistent positioning of one eye relative to the center of the portrait is not mentioned in art texts.

Fig. 1 To illustrate the degree to which an eye tends to be set on the center vertical in portraits, twelve classic portraits from the past five centuries were selected for reproduction. Selection for this figure was based on depiction of one eye as compositionally dominant over the other with the head in a variety of poses, with no attempt at a scientific sampling procedure. Each portrait is reproduced in its full width, but many are cropped in height to fit into the panels. From the perspective of frame geometry, many of the examples illustrate the lengths to which portrait artists seem to go to set the dominant eye on the center line. Several cases illustrate how clever composition generates the overall impression that the face is symmetrically located in the frame. Only when the guide lines are drawn through the pictures does it become clear that one eye is at the exact horizontal center.

Fig. 2. Eye strips of a sequential series of 21 classical portraits from Schneider's The Portrait to provide a visual depiction of the eye placement. Each strip includes the full width of the painting from which it came, with a range of about 2:1 in width over the set of paintings; the widths were all normalized to the same value to provide an evaluation in terms of frame geometry. Note that about 2/3 of the eyes have one of the pair positioned within the range of +5% of the distribution (white lines), which is about one eye width.


Compositional principles that may be hypothesized as the basis for eye placement:

Fig. 3a. Major axis hypotheses. These axes intersect at the center of the frame, generating the competing hypotheses that either the center of symmetry or the closer eye should be centered at the center of the frame. The most important of the four axes is the vertical, so another hypothesis is that one of the two eye parameters would be commonly located near the vertical axis. Conversely, if a preference for compositional asymmetry would lead to avoidance of the center vertical, one of the two eye parameters might be positioned close to one of the diagonals in the frame, or along the horizontal line.

Fig.3b.Golden Section hypotheses. The Golden Section, in which the smaller subdivision of a line is to the larger as the larger subdivision is to the whole line, is commonly cited as an aesthetic principle. The four dashed lines each divide the frame according to the Golden Section ratio of 0.618. The central square and its corners represent the predicted eye placement loci if it were guided by the principle of the Golden Section.

Fig. 3c. Head-centered hypothesis. If the head is centered in the frame, the mean positions of the two eyes will fall symmetrically around the center. The base of the triangle is specified according to the assumptions that the head will always fit inside the frame, the eyes are half-way up the head and the distance between the eyes is half the head width. The angled sides are based on the assumption that the head is reduced in size according to the amount o the body that is depicted, which pushes the head up to near the top of the picture.

Fig. 3d. Eye-centering hypothesis. A compositional principle developed post-hoc from the data analyzed in this paper, that one eye tends to be placed at the lateral center of the frame. In this case, the positions of the other eye would fall in the form of an inverted V on either side of the center.


Fig. 4. Distributions of the measurements of portraits from 282 different artists, for the position of the best-centered eye and the center of symmetry ('mean binocular'). The distribution for the best-centered eye (solid line) is centered in a tight distribution with a standard deviation of about +/- 5% around the vertical axis. An alternative hypothesis is that the face rather than just one eye is being centered in the frame. An estimate of the face position is its center of symmetry, given by the halfway point between the two eyes (labeled 'mean binocular'). The distribution for this mean binocular point (dashed curve) shows a bimodal distribution, implying that the center of symmetry of the face is NOT being accurately centered in the frame. Instead it is placed either on one or the other of the center line as expected if it was one or other eye of the pair that was being centered.

A second question is how the eyes are distributed over the two-dimensional space of the canvas. The eyes in portraits tend to cluster horizontally around the center vertical, with one eye centered in a normal distribution with a of only + 5% of the frame width (full curve; second eyes omitted for clarity). The binocular mean (dashed curve) had a bimodal distribution implying that one or other eye was usually centered. Conversely, the eye height distribution was not centered vertically but peaked (as eye icon) close to the classic Golden Ratio of 0.618 (where the smaller portion has the same ratio to the larger as does the larger to the whole), with virtually no eyes below the vertical center.

Fig. 5. The lateral positions of each eye as a proportion of frame width are plotted against verical position (only the area above the center of the canvas is shown in the graph) for portraits from 282 different artists. Filled symbols: most centered eye; open symbols: other eye (circles: leftward eye; triangles: rightward eye). Note that the position of the most centered eye (filled symbols) forms a narrow vertical strip of about 10% of the frame width regardless of vertical position (which tends to vary with head size). The centered disribution is astonishingly narrow and symmetrical considering that the images are drawn from diverse cultures over five centuries. The distribution of the 'other' eyes matches the tripod-shaped distribution of the pure eye-centering hypothesis and effectively discounts the other three hypoetheses.


The eye centering with an accuracy of ~1 eye width is barely mentioned in art criticism, suggesting that unconscious functions operate in our aesthetic judgements.

This page last updated 12/27/00

Go to...
ARVO 97 The Eye Page Symmetry Book Tyler lab