I haven't personally experimented with the reliability or validity of any kind of pedometer, but that's where research studies can help

. Interpreted with caution, of course.
I looked at three recent North American studies (done by practically all the same researchers, so perhaps researcher bias?) and it seems that agreement was in favour of Kenz Lifecorder, New-Lifestyles NL-2000, and Yamax Digi-Walker SW-701 for being the most accurate in assessing actual number of steps (within 3%).
There were actually 14 models that were tested among the 3 studies I looked at -- all the pedometers worked best at a pace of 80+ m/min (they underestimated steps at 54 m or less/min and in those with gait disorders).
In other outcome measures, none of the pedometers fared too well on calculating the distance walked, and were least reliable in predicting the number of calories burned.
Just for interest, this was what I gathered (# of studies model was in, accuracy in assessing steps):
Accusplit Alliance 1510 (1, Underestimated)
Freestyle Pacer Pro (3, Underestimated)
Colorado on the Move (1, Underestimated)
Kenz Lifecorder (3)
New-Lifestyles NL-2000 (3)
Omron HJ-105 (2, Overestimated, Least accurate)
Oregon Scientific PE316CA (3, Overestimated)
Sportline 330 (3, Least accurate)
Sportline 345 (2, Underestimated)
Walk4Life LS 2525 (3, Overestimated)
Yamax Skeletone EM-180 (1, Underestimated)
Yamax Digi-Walker SW-200 (2)
Yamax Digi-Walker SW-701 (3)
Yamax Skeletone EM-180 (2)
I also stumbled upon info regarding Global Positioning System (GPS), or satellite positioning, technology in studying gait biomechanics -- very interesting!