How can I set up nonlinear regression to constrain a parameter so that its best fit value must be greater than 0.

WIth Prism 4 and 5 (and later) , this is easy. Just go to  the "Constraints" tab of the nonlinear regression dialog, and set your constraint.

Prism 3 does not include an automatic way to include constraints. But you can add constraints indirectly. Use the ABS()  function, so that a parameter enters the equation as its absolute value. Here is an example, using the equation for two-site binding and constraining both Kd values to be positive. 

Note that the equation does not really constrain the best-fit value of Kd1 and Kd2 to be positive. What it does is make sure the effective value of the parameters in the equation is positive, even if the parameter ends up with a negative value. So if the best-fit results are negative, just "white out" the minus sign, knowing that the effective value in the equation was positive.