During Nonlinear optimization using lsqnonlin(), is it desired to have resnorm value as low as possible?

I have been using lsqnonlin()for my data. I have put different initial guess to initiate the process and also scaled it since my output supposed to be very low in the order of 10^(-12) or lower. Different scaling,different initial guess gives different output. Also tried to do this using the final output from first run as a next initial guess. But, Every time I get 'local minimum possible' but getting different resnorm values. What should be final stopping criteria where I could say that it could be final solution?