The screws, of course, were made with a great degree of accuracy and as free as possible from any unevenness. Having tested screw _a_, I next tested screw _b_. I found with the same number of revolutions per minute that this screw produced more thrust, but it required more power to run it, and when the energy imparted to the air was compared with the readings of the dynamometer, it was found that it did not do quite so well as _a_; still as the thrust was greater and the efficiency only slightly less, it appeared to be the better screw. Upon trying screw _c_, under the same conditions, the thrust was very much increased, but the power required was also increased to a still greater degree, showing that this form was not so favourable as either _a_ or _b_. All the screws experimented with had very thin blades, and it occurred to me that the difference between _a_ and _b_ might arise from the fact that, when _a_ was running at a very high velocity, the working side instead of being flat might have become convex to a slight extent, whereas with _b_, a slight bending back of the edges of the blade would still leave the working side concave. I therefore made the screw shown at _e_, which had the same pitch as the other three, but the working side was of the same shape as _a_. Of course the additional thickness of the blades made it impossible to give an easy curve to the back. Curiously enough I found that _e_ did nearly as well as _a_, and quite as well as _b_. The additional thickness did not interfere to any appreciable extent with its efficiency. I then made another propeller, shown at _g_, which was of the same thickness in the middle as _e_. Upon running this, I found that it required considerable power, and no matter which way it was run, the thrust was always in the direction of the convex side, which was quite the reverse from what one would have naturally supposed. [Illustration: Fig. 15.--The manner of building up the large screws.]