Thanks to both of you, Andrew & Bruce.

I made my own research regarding Median, asked my colleague to bring statistical book. She brought 'Introduction to Mathematical Statistics' 3rd edition by R.V. Hogg/A.T. Craig. The definition of Median there is quite complicated:
A median of a distribution of one random variable X of the discrete or continuous type is a value of x such that Pr(X=1/2. If there is only one such x, it's called the median of the distribution.

Anyway, couple of people also confirmed the definition given by Andrew Coates.
Therefore I would use this definition in my calculations. I'll get rid of the limitation by number of records.

BTW, yesterday I created a FAQ called GetMedian. I'll correct it today.

>>Actually, this restriction gave me my manager. It's probably incorrect too, since his definition of Median function was wrong.
>
>Andrew is right, your manager is wrong. Even one point has a median, though not highly useful. Certainly 2 points have a useful median, and it's equal to the mean.
>
>That's spoken statistically. Practically speaking, there may be some reason why a small number of points is not a sufficient number for computing for some reason...there is, for example, a problem with variance and degrees of freedom in estimation when there are very few data points.