My advice would vary, depending on the exact particulars of what you're trying to do, but I would think that, generally, you could collapse the repeated measurements by taking their mean or median at each ordinate, and fit a curve to this averaged data.