This was originally written in answer to a question postted on the Automotive Body Engineering forum, related to a Geo wtih 3-cylinder (!) engine, but most of the remarks have general application:
I'd echo what Greg Locock says: the next book you find on automotive NVH will be the first one!
You can do some work inexpensively with a laptop, with the built-in "Sound recorder" application and using an inexpensive Radio Shack microphone, if you can work with comparative results only, since you won't be able to calibrate the input level on a .wav file.
You can find a freeware version of CoolEdit software, something like "CoolEdit 97" (or is it "96") that is available on the web; it has a good analysis function that will permit doing FFT analyses. This won't tell you how loud it is inside your Geo (I suspect that answer to be "very") but you will be able to determine the relatively highest level frequencies, and comparison testing should give you an indication of whether or not you have reduced those peaks.
A bit of detective work will tell you the source of the various peak frequencies, more easily if you construct an "alignment diagram," which is a plot of frequency vs engine RPM on which you have plotted the lines showing engine firing frequency, alternator rotation and firing frequencies, and any other roational inputs you may have.
For example, your 3-cylinder engine fires 3 times for every two rotations of the crankshaft, and so is called a 1.5-order phenomenon: at for example 1500 rpm, engine firing is (1500/60)*(1.5) = 37.5 Hz
So your engine firing is a line on the alignment diagram from the origin and passing through the point (37.5 Hz, 1500 rpm). Alternator would be a line at the alternator drive ratio, e.g., 2.5 x engine speed perhaps, and another line at 3 x that, owing to the nature of alternators. That is, at the same 1500 rpm, alternator rotation might be 1500/60 * 3 = 75 Hz, and alternator "firing" would be at 75 * 3 = 225 Hz.
You'll have to decide on a repeatable test procedure, someplace where you can drive a steady speed at a constant load, without distractions, excessive noise from passing cars, with a uniform level road surface - in fact, if you find such a place let me know where it is... You'll also have to make sure your test conditions are well-documented: wind speed, ambient temperature, etc., and make sure the car is fully-warmed-up. Small variations can sometimes make large differences in noise.
Now, when you detect a noise peak in the frequency spectrum, put its frequency/rpm point on the alignment diagram, and with luck, it'll land on an existing line, and point you in the direction of the problem.
In order for this to work, of course, you'll need a fairly accurate tachometer; you should be able to check the accuracy of your tach by finding the firing-frequency spike in the noise spectrum (rpm = freq * 40, as above) and comparing it to what the
Simply knowing what the source is doesn't always suggest the solution: for example, say you know that engine firing is the problem at highway speed: is the noise coming from the muffler, from the aircleaner, or in a car this old, via bad engine mounts, or an exhaust system that's contacting the chassis? More detective work is required, and space and my time prohibits me from describing all the common techniques for doing this.
One more point: spray on stuff looks great, but may or may not do any good. For your interior, try to maximize the amount of carpet and fabric upholstery, preferably backed with thick open-cell foam.
Heavy vinyl barrier sheet or (better yet) a thin sheet of lead under the carpet will make an effective sound barrier.
