Based on your question, I am assuming that you are trying to fit a discrete curve. This is a hard problem in general but you can obtain a reasonable solution by using the "lsqnonlin" function in the Optimization Toolbox. More information can be found here:
In the context of least squares, residual is a more sensible metric than standard deviation. You could calculate the R^2 metric for the fit using the residual returned from "lsqnonlin".
