How to test how well data points fit a curve

I have generated a 1x100 prediction based on a set of observations. The prediction is in the form of a curve. I would like to quantify how well the prediction fits a seperate set of observations (different than the one I generated the predictions from). Here's a visual representation (two curves are shown):

What might be the best way to quantify the predictive accuracy of each model with respect to the observed data (black diamonds)?

x = Contrast is a 1 x 100 vector

y1 = Linear Summation is a 1x100 vector

y2 = Binocular data is a 4x1 vector

Answer 1

[Image Analyst]

How about the mean or median of absolute differences? Or mean square error, or sum of residuals, or RMSE, or correlation coefficient. What metric would you LIKE to use to describe the difference?

[Blake Mitchell]

I calculated RMSE but I was wondering what the best practice might be. Do you think it would be better to fit a line through the observed data points first and then compared the model to the fitted observed? And if so, would it come down to just quantifying the difference between the two curves?

[Image Analyst]

Not sure what you're going to do with the result. Maybe it doesn't matter and RMSE is just as good as any of them. Just go with that and see if it works for you.