Whenever doing calibration, you will be confronted with some non-uniqueness of the solution. In the simplest case, you might not know whether you need to decrease recharge or increase conductivity. If confindence levels on both sets of input data are equally low and head observations are the only calibration targets, there is no way to do the 'right' thing. So either you manage to come up with additional calibration target or parameter data, or you need to at least qualitatively think about what it would mean for your results (in terms of decision support done with the model, not in terms of resulting hydraulic heads!) if your decision was wrong. Going further, fully-blown uncertainty analysis (e.g., applying PEST) could help to identify parameter correlations and result uncertainty.