It depends on what you are trying to prove. If you reckon that the subject
is likely to be over (or under) the average then that is your hypothesis and
you will go for a one tail test. At 95% confidence level that means that the
subject will only be less than that in 1 out of twenty times. The 5% must all
be in one tail of the Normal curve.
In a two tailed test the results can be either above or below the average.
95% confidence is split 2½% below and 2½ above.