Stephenson:Neal:Quicksilver:222:The sun becomes an oval (John B.)
From the Quicksilver Metaweb.
I think more needs to be said about the illusion that the sun, and the easier-to-observe moon, have a different size or shape when viewed at the horizon. It is not true that atmospheric refraction plays the most significant role (of course it plays some). While the sun and moon do become ovals at the horizon, more noticeable is their change in apparent size, which is often attributed to atmospheric distortion.
This is a very involved issue, and the best explanations are to be found at http://www.lhup.edu/~dsimanek/3d/moonillu.htm
There are several optical illusions (especially the Ponzo Illusion) that are invoked to explain why the sun and moon appear to have different sizes at different places in the sky. Here is one explanation. The sky is a hemisphere above you, a perfect half of a sphere. And yet it is perceived as more of a bowl. The top of the sky seems less distant than the sky at the horizon. However, this is not so.
The moon at its zenith is (roughly) as far from an observer on earth as it is at the horizon. But as its position in the sky is apparently farther from earth, people perceive it as larger.
This explanation, though common, has holes. The so-called "Moon Illusion" seems to have as much to do with human perception as with actual amount of sky taken up.
The above is not entirely accurate. A ray from a point on the surface of the earth extending straight up passes through less atmosphere than a ray from the same point passing over the horizon. This is clear as the thickness of the atmosphere is very small compared to the radius of the earth. The linked page says as much, "At the horizon the light must pass through a greater distance in our atmosphere than when the moon is higher in the sky."
Yes, and the refraction effect does take place through a convex shaped surface (the upper bounds of the atmosphere curving around the surface of the earth). The refraction effect that shortens the height of the moon at the horizon is a case of direction linear refraction. The moon is also widened by the horizontal cross section of the atmosphere of the horizon which describes a convex lense. At the horizon, the moon is 4,000 miles further from the observer than it is when directly overhead, which is a difference of only 2%, but is discernable, but is overpowered by the horizontal and vertical refractive effects of the spherical atmosphere.