So you ask, how do you find out what a good shape actually is? Fortunately there are some tools available on the web. The Applied Aero site has a number of java applets that allow you to interactively see what shapes generate the properties you desire. In this case what we are looking for is the lowest most evenly distributed pressure over a symmetrical body. I was able to model this with their Airfoil pressure distribution java applet. This applet allows you to change wing shapes, and immediately see the pressure distributions. Below is the same NACA optimal wing section as above, showing the pressure gradients. Note that pressures decrease toward the top of the graphic. In the picture below, there is a high pressure spike at the leading edge, sharply transitioning to a very low pressure area, then the pressure slowly rises toward the tail of the wing section. Though this would not make a bad fairing, we can do better. 

Ideally we would have a shape that has no large variations in pressure along the body of the fairing, have the pressure drop slightly as it travels along the fairing, and then increase (not too quickly) at the tail. This will allow the layer of air that flows along the surface of the shape, called the "boundary layer", to remain attached to the shape, and to be sucked toward the rear of the fairing. This is called laminar flow.  A shape that is highly laminar will be very fast. The idea here is to get a close to this shape as possible. Let's try reducing that large low pressure zone in front. As you can see by the low pressure bump, followed by a sharp increase in pressure toward the rear of the shape below, now the tail is too short. While the picture above will have the boundary layer delaminating near the nose, this scenario will cause the boundary layer air to delaminate near the tail, slowing you down.

Faster Shapes!
Though not ideal, the shape below looks better. The pressure is fairly evenly distributed, and the pressure increases slowly at the tail.  As an added bonus, it looks like a shape that a human could fit in it...

To corroborate my thoughts that the shape above was close to optimum, I found that there is something called the "optimum body of revolution", which the picture below illustrates. Note that the Cp (pressure) curve is relatively flat for the entire length of the shape, and drops slightly before rising at the tail. This is a good thing. This shape looks a lot like Terry Hreno's "Moby" streamliner design.

Here's a one of the potential fairing shapes that I was working with. It's very close to the "optimum body of rotation" illustrated above.  Click on the picture below for a higher resolution bitmap

The next part was deciding on the shape from the side. This is trickier, because the aeronautics folks don't model anything that cruises 3" off the ground, and the automotive folks don't care about aerodynamics that much. What I have learned from practical experience says that you want to minimize the pressure that builds up under the fairing. This means you want your fairing to mostly deflect the air to the sides, and up over the top of the fairing, instead of down and under it. If you design the fairing with a symmetrical nose from a side view, air will be forced under the bike, causing a venturi effect which constricts the airflow, and slows you down. Of course if you are building a fairing that will be higher above the ground, this is not so much of a factor. Back to the pressure graphs. Here's a shape with a flat bottom, and a bulging top. Note that the pressure on the bottom surface is very fairly neutral, while the top has a big high pressure spike followed by a sharp transition to a very low pressure, which continues along the fairing until the tail, then sharply rises again. These sharp gradients will cause the air to delaminate, which raises the drag. On the real 3D fairing much of the air will be deflected to the sides, rather than over the top, reducing these radical pressure variants.

Strangely, this is close to the shape that the aeronautics folks use for a "business craft fuselage" as represented in the graphic below. Note that the pressure distribution along the bottom is near zero, and along the top is not too bad, due to the gradual lead in. This is the side view shape I decided to use for the front of the fairing. The back of the fairing will need to be fairly squared off, due to the vertical shape of it's human payload, and my desire to keep the length under 10 feet.

Cutting Foam:
Here's the template that I used to create the Barracuda fairing (top view).   I started by making a full sized cardboard template of the shape detailed below (you can click on it for a higher resolution picture). I then cut a stack of 2" sheet Styrofoam insulation (the kind you find at the builders supply store), one at a time.. As my fairing is about 30" high, I cut 15 identical shapes, and stacked them. I glued them together with copious amounts of 3M super 77 spray adhesive.

The side view was a little more free form. I measured my footbox and knee clearance, and by more or less following the "business craft fuselage" form, drew the side fairing shape.

As I decided to build the front end first and then worry about the tail later, the tail is not detailed in this drawing. I basically just ended up following the contours and rounding everything off to complete the tail. You can click on the drawing below to see the details of the template design. To cut this contour in the stacked block of Styrofoam, we used a hotwire. A hotwire can be easily made by stringing a stainless steel wire between a two metal rods, which are inserted into holes in a 2x4. Connecting the leads of a 10 amp car battery charger to the ends of the hot wire causes it to heat up enough to melt the Styrofoam quite handily. Be sure to keep the wire tight!

Here's a side view with foam elevations:

Once the basic shape was cut from the foam, it was basically a matter of rounding the form with a big file until the shape was aesthetically correct. As the old adage goes, fair to the eye is fair to the air!

Building the Barracuda fairing

Other shapes:
Fairing shapes are as varied as car bodies. There is no "right answer" when it comes to designing a fairing. Many factors are involved, all involving trade offs. If you want a medium speed fairing, that handles well in the wind, you'll want to keep it as short as possible, and round the sides as much as possible to allow the wind to flow around with less of a sail effect. There is also the the fast "hatchet" fairing design, where the body is kept short and extremely narrow, with sharply tapered nose and tail. Other fairing designs are made to fit particular bikes.

Thanks to Matt Weaver for his help in clarifying some of my thoughts and assumptions above. Matt also forwarded some very helpful Aerodynamics notes by Mark Drela, builder of the world distance record "Deadalus" human-powered airplane, and the "Decavitator" world speed-record human-powered hydrofoil boat, which he kindly extracted from the 1993 HPVA archives.

Some Aerodynamics Notes by Mark Dreyla:

Q:
What are some of the general guidelines of fairing design for objects moving in the HPV speed domain? 

A:
The theoretical difference between the drag of a sharp vs. rounded nose are miniscule. I'd say the blunt nose is preferable since it gives a shorter fairing, and is safer around pedestrians. However, when I say "rounded" I do not mean hemispherical! A hemispherical nose on a cylindrical body has rather high drag compared to a 3:1 elliptical nose, say. The best fairing shape significantly depends on whether you can achieve extensive laminar flow. By "extensive" I mean over 50% of the body or more. It's easy to get laminar flow on the front 10%, but this doesn't help much to reduce drag. If going after extensive laminar flow, the shape -- and surface curvatures in particular -- need to be precisely controlled over the front 40-60% of the body. Most likely, only a high-quality fully-molded fairing with no transverse seams on the front half would be good enough. Unless the shape is very close to a body of revolution, the shape will likely have to be designed with fairly sophisticated methods to tailor the surface pressure distributions. A further difficulty is that the environment on an HPV is not conducive to extensive laminar flow. Water beads, mud splotches, and road vibrations in the 50-500 Hz range all tend to kill laminar flow. A straight swept leading edge, like what might result from a flat bent windshield, also tends to be destructive to laminar flow. If most of the flow is turbulent, then the exact shape of the fairing will have very little effect. Trying to stack precise NACA airfoil shapes is largely a waste of time in this case. Wrapping a smoothly-flowing shape with a near-elliptical nose will be as good if not better than trying the force-fit standard airfoil shapes. 

Some basic "rules" are: 

• Avoid outside corners and areas of sharp streamwise curvature. Shaving off a few square inches of frontal area by carving a flat spot at the widest point will only increase the drag. 

• Avoid sudden changes in streamwise curvature. The hemispherical nose butting straight into a zero-curvature cylindrical section is a good example of a bad shape. There will usually be a large velocity spike near the curvature discontinuity, followed by a small separation zone.

• Avoid aft-facing steps and air leaks at and behind the maximum-thickness point. The small recirculating zone behind a step might not be able to close in the aft pressure rise and might precipitate large-scale flow separation. Leaking "dead" air into the boundary layer over the pressure rise is very detrimental for the same reason. Minimizing the flow out of wheel cutouts is important. The best place to dump ventilation air flow is out of trailing edge.

• The back edge or point does not have to be perfectly sharp. A good rule of thumb is that if the area of the flat base is less than drag_area/4, there will be no drag penalty. A good HPV fairing might have CD = 0.10 based on a frontal area of 5 ft^2, in which case the allowable base area would be 0.10 x 5 / 4 = 0.125 ft^2. This might make the fairing a bit shorter. 

• Surface finish is important, but there is a threshold roughness height k_min below which there is no further drag reduction. This is given by k_min ~ 4 sqrt(2/Cf) nu / V nu = kinematic air viscosity = 1.4e-5 m^2/s V = local flow velocity (m/s) Cf = local skin friction coefficient ~ 0.025 (nu/VL)^(1/7) L = distance from leading edge (m) For a typical fairing at V = 10 m/s this works out to Cf ~ 0.004, k_min ~ 0.1 mm = 0.005 in, which is not all that smooth -- like the surface of galvanized steel, say. A doped cloth surface is probably close to this threshold. For a speed HPV this roughness threshold would be quite smaller.

As far as "truck suck" or crosswind sensitivity, there is very little that can be done other than reduce the height and side area as much as possible. The "airfoil" shape of the fairing will have little effect.

Also check out: http://www.princeton.edu/~asmits/Bicycle_web/blunt.html