False detection of voicing


I have a collection of sound files recorded in a quiet room. While analyzing them, I noticed that Praat often detected voicing (i.e., a visible pitch curve) in what I would consider to be voiceless consonants. I played with the pitch settings in attempt to get rid of these "false" sections but was unable to (I managed to remove the pitch curve from vowels but not from the voiceless consonants).
Interestingly, when I changed the maximum value of the pitch range, the pitch value in truly voiced sections remained relatively constant, but the value of the false pitch curves always shifted upwards and remained close to the upper bound of the range.
What can I do to get a more reliable pitch analysis? (the false pitches have a large and undesired effect on mean pitch calculations)