Let the sample size be N, then the standard error of the sample mean will

be approximatly 10/sqrt(N) minutes. A 90% interval (assuming a normal

distribution on the mean, which will not be too bad an assumption assuming

N is reasonably large - often as few as 12 will be adequate to make this a

reasonable assumption) is +/-1.645 SE's about the mean, so the width of

this is 3.29 SE's. So we want:

3.29 x 10/sqrt(N) <= 5 minutes,

so:

sqrt(N) >= 3.29 x 2 = 6.58,

or:

N >= 43.30

So the smallest sample size that will be this will be 44.

RonL