Start by assuming its a quadratic:

y=ax^2+bx+c

Sort the data points into scending order for x.

Put the values for the first third and fifth data points into this which will give you a set of three simultaneous equations for this, which you can solve for a,b and c.

Alternativly look up quadratic regression on Google.

If you don't want a quadratic fit you can repeat the first method for a cubic or quartic.

It would also help if you provided more background information on the source of this question.

CB