From what little I understand about it, "lie detectors" most commonly measure levels of skin conductivity or GSR (galvanic skin response). This changes as your "psychological arousal" changes. Generally, when people lie, they become somewhat more tense/emotionally active, and the changes in electrical conductivity can be measured.
Problem is, being nervous or startled by a question will also lead to this, so a "lie" detector doesn't actually measure whetehr you lie - it measure something that happens when you lie, but that also happens when you do other things.
I only know this because I am looking into measure stuff like this as part of my ongoing research. Pupil diameter is another thing that can be measured that is a good proxy for things like arousal and cognitive load. ... Soon, I'll be like Harrison Ford is Blade Runner - hooking students up to machines and playing mind games with them. MUWHAHAHAHAHA! [:D]