I have looked at the documentation, but it does not spell very clearly what the objective function for NonlinearModelFit is. It does say that it will return the same result as FindFit, which by default returns a least squares fit.

In FindFit, the objective function (the function that is minimized) can be changed by using the option NormFunction. It seems like in NonlinearModelFit there is no such option, so one is limited to whatever can be achieved by entering weights or doing transformations on the data. Say I wanted to find a fit that optimized yObserved/yPredicted (e.g., by minimizing the sum of (yObserved/yPredicted -1)^2 ). It looks like the only way to achieve this would be to take logs of my model form and logs of the observed data, but that seems rather clumsy.

Any ideas?

Thanks, OL.