The threshold for differentiation was computed from the data on the 
basis of the convention that the threshold was represented by 50% 
discrimination. 
A reflected circle will appear as a circle so long as it is 
viewed along a line disposed to the normal at an angle equal to that 
of the incident ray normal to the object. It is the permissible de 
parture from this condition that is sought as the 'critical viewing 
angle'. 
The experiments for the visual angles 5, 15, and 20 minutes 
yielded ellipticity thresholds of 10" ±, 3", 14" ±. 3", and 18" ±. 3", 
and corresponding critical viewing angles of 15° ±, 2°15', 10° ± 1°, 
and 10° ±. 1° respectively, where the ranges indicated are standard 
deviations. Simple alignment procedures were consequently used. 
2. The Experimental Procedure 
The desire for objectivity cannot be fully satisfied in in 
vestigations of this type because the response of a subject is in 
volved. Any threshold has to be measured repeatedly and an average 
found, because it may vary from moment to moment, and may be ob 
scured by its variability in time. 
It was decided to use the method of adjustment, or repeated 
setting, rather than the 'constant' methods because it is the method 
used in photogrammetric practice, it is more amenable to statistical 
treatment, and the final judgement and setting requires direct par 
ticipation by the observer. 
Numerous experimental studies (1) (2) (6) (13) have demonstrated 
a significant relationship between training or practice and a variety 
of perceptual functions. Practice produces an improvement in the pre 
cision of perceptual judgements, and this was most noticeable in these 
observations. It was extremely difficult to tell when the effects of 
practice began to influence the results, but, after inspection of the 
observations, an arbitrary decision was made to reject all obser 
vations taken during an initial three-week period. 
In the design of the observation series, a number had to be 
chosen sufficiently large to permit adequate statistical treatment, 
yet not so large as to introduce the deleterious affects of fatigue 
towards the end of a series. A series of 50 observations was taken 
as the basic unit, with a trial set based on a series of 10 for com 
parison. Generally, each point determined was based on the mean of