In nonlinear regression, how can I constrain the sum of two parameters to be less than (or greater than) some value?

Prism does not allow you to constrain the sum of two parameters. But you can get the effect you want by rewriting the equation. This is best seen through an example. Say you want to fit the two phase exponential association equation, but constraint the sum of the two Ymax values to be less than 1.0. The equation for the model is

 Y=Ymax1*(1-exp(-K1*X))+Ymax2*(1-exp(-K2*X)).

You want to contrain Ymax2+Ymax2 to be less than 1.0, but Prism does not let you do this. Instead rewrite the equation like this:

  Ymax1=YmaxTotal * F1
 Ymax2=YmaxTotal - Ymax1
 Y=Ymax1*(1-exp(-K1*X))+Ymax2*(1-exp(-K2*X))

So now you are not fitting Ymax1 and Ymax2, as these are intermediate variables. Instead you are fitting YmaxTotal and F1, which is the fraction of the total that comprises the first phase. WIth Prism 4, you can contrain YmaxTotal to be less than 1.0, and constrain F1 to be between 0 and 1.