A ray of sunlight falling on a prism of glass or crystal does not emerge unchanged in character. Different portions of the ray are differently bent, so that when they emerge from the prism they no longer travel side by side as before. The violet part of the light is bent most, the red least; the various colors from violet through blue, green, and yellow, to red being bent gradually less and less. The prism then _sorts_, or _sifts_, the light-waves. But we want the means of sifting the light-waves more thoroughly. The reader must bear with me while I describe, as exactly as possible in the brief space available to me, the way in which the first rough work of the prism has been modified into the delicate and significant work of the spectroscope. It is well worth while to form clear views on this point, because so many of the wonders of modern science are associated with spectroscopic analysis. If, through a small round hole in a shutter, light is admitted into a darkened room, and a prism be placed with its refracting angle downward and horizontal, a vertical spectrum, having its violet end uppermost, will be formed on a screen suitably placed to receive it. But now let us consider what this spectrum really is. If we take the light-waves corresponding to any particular color, we know, from optical considerations, that these waves emerge from the prism in a pencil exactly resembling in shape the pencil of white light which falls on the prism. They therefore form a small circular or oval image on their own proper part of the spectrum. Hence the spectrum is in reality formed of a multitude of overlapping images, varying in color from violet to red. It thus appears as a rainbow-tinted streak, presenting every gradation of color between the utmost limits of visibility at the violet and red extremities. If we had a square aperture to admit the light, we should get a similar result. If the aperture were oblong, there would still be overlapping images; but if the length of the oblong were horizontal, then, since each image would also be a horizontally placed oblong, the overlapping would be less than when the images were square. Suppose we diminish the overlapping as much as possible? in other words, suppose we make the oblong slit as narrow as possible? Then, unless there were in reality an infinite number of images distributed all along the spectrum from top to bottom, the images might be so narrowed as not to overlap; in which case, of course, there would be horizontal dark spaces or gaps in our spectrum. Or, again, if we failed in finding gaps of this sort by simply narrowing the aperture, we might lengthen the spectrum by increasing the refracting angle of the prism, or by using several prisms, and so on.