Nonlinear regression works iteratively. Prism starts with initial estimated values for each parameter. It then gradually adjusts these until it converges on the best fit. "Converged" means that any small change in parameter values creates a curve that fits worse (higher sum-of-squares). But in some cases, it simply can't converge on a best fit, and gives up with the message 'not converged'. This happens in two situations:
•The model simply doesn't fit the data very well. Perhaps you picked the wrong model, or applied the wrong constraints.
•The initial values generated a curve that didn't come close to the points. In that case, Prism may not be able to figure out how to change the parameters to make the curve fit well.
Did you enter the right model?
Does the curve defined by your initial values come near your data? Check the option on the diagnostics tab to plot that curve.
If you entered constraints, were they entered correctly?
If you didn't enter any constraints, consider whether you can constrain one or more parameters to a constant value? For example, in a dose-response curve can you constrain the bottom plateau to be zero?
Can you share a parameter over all the data sets (global fitting)?