The test tones should not be harmonically related. Most commonly used are 700 & 1900 Hz (ARRL) and 800 & 1800 (some commercial/military).
A sound card should be capable of putting out two clean tones, but if you prefer a stand-alone device, Elecraft manufactures an moderately priced two-tone test generator kit for 700 & 1900 Hz, model 2T-gen for US$ 69.95.
The old Motorola application note EB38 may also be worth reading:
http://www.freescale.com/files/rf_if/doc/eng_bulletin/EB38.pdfThe ARRL's test procedure manual is also good reading -
http://www.arrl.org/files/file/Technology/tis/info/pdf/Procedure%20Manual%202010%20with%20page%20breaks.pdfYou should also be aware that there are two ways of defining the intermodulation level. The commercial / military standard references the output intermodulation product to the power of a single tone. This is easily read on the spectrum analyzer as the difference in dB between one test tone and the IMD product. The ARRL method references the intermodulation product to PEP, which is 6 dB above the single tone. If you want ARRL protocol, make the same measurement, but add 6 dB to the difference.
As an example, if the 3rd order IMD product is 30 dB below one of the reference tones, the commercial/military specification will be "IMD product 30 dB down" and the ARRL's version will be "IMD product 36 dB down."
I prefer the commercial/standard methodology, but either is OK, provided you make it clear which reference is being used. If you wish to compare your measurements with ARRL data, then keep this 6 dB difference in mind.