central limit theorem

[stopgo]

This seems wrong, but I don't understand why.  If I have a wildly inaccurate sensor, but one which gave random error within a certain range, and I take a large (approaching infinite) set of sensor readings and then average them, shouldn't the value I get approach the true mean?

[Admin]

I take a large (approaching infinite) set of sensor readings and then average them, shouldn't the value I get approach the true mean?

Yes, you will get the 'true mean.' But you are in no way guaranteed that this mean is going to be the correct value, since error could be biased.

You should hook up your sensor to an oscope/hyperterminal and figure out what the error looks like, where it comes from, how common it is, etc. Some sensors have predictable error, in which you can just filter out the bogus results.