There are some interesting threads on the email@example.com list about CW. The question came up "What is CW?", both as to technology (how is it generated) and regulation (how is it defined.) Here are my 2 cents on the subject.
(If you're new to ham radio, CW stands for "continuous wave". In traditional Morse radiotelegraphy, your transmitter sends a steady wave when your key is down and no signal when your key is up.)
Having nothing better to do, I went to the FCC website to read up on Part 97 regulations and what they say about CW. The relevant sections are 97.3 which refer back to 2.201 and 2.202. Some excerpts are at the end. Classical amateur CW might be 100HA1A, specifying 100 Hz bandwidth, or simply A1A. The ARRL FCC Rule Book has some useful material, too.
It seems that the FCC is interested in the signal that shows up on the air and not how it is generated. Fair enough. Normal amateur CW is A1A, I believe. Some generation methods (like audio tones into a not-so-good SSB rig) are worse than others. FCC requires signal purity to observe good engineering practices, or words to that effect, and that may rule out the KWM-1 technique nowadays. The DSP method (e.g., Ten-Tec Orion) can be as perfect as you're willing to pay for.
As the Rule Book (8th ed.) explains, it would be possible to narrow the "100 Hz" DSB spectrum of an A1A signal by eliminating one sideband (50 Hz) and suppressing the carrier. (However you make it, CW does have a carrier and sidebands just like a voice signal.*) I wonder if anyone has ever done it, and whether a half-width carrier-less CW (or psk31?) signal would be decodeable after HF propagation. You'd need really tight frequency and passband control.
*A carrier? What about between characters? Yes, mathematically the carrier is still there -- even after you turn your rig off. Of course, there are also very low freq sidebands that conveniently cancel out the voltage... So your rig had better be very very linear or it won't be safe to shut off the power! Don't lose sleep over it.